System, method, and computer program product for improved model training using reinforcement learning and feature dropout

Abstract

Systems, methods, and computer program products are provided for improved model training using reinforcement learning and feature dropout. An example system includes at least one processor configured to receive a pretrained machine learning model, which includes a first input layer and first hidden layers and has been trained based on training data including data records, each including a plurality of features. A second machine learning model is initialized and includes a second input layer, a feature selection layer, and second hidden layers. The second input layer and second hidden layers are initialized based on the first input layer and the first plurality of hidden layers, respectively. The feature selection layer includes weights including a respective weight for each respective feature. Each weight is associated with a dropout rate for the respective feature. The weights are adjusted based on the training data and at least one reinforcement learning reward.

Description

Attorney Docket No. 08223-2405237 (8270W001 )SYSTEM, METHOD, AND COMPUTER PROGRAM PRODUCT FOR IMPROVED MODEL TRAINING USING REINFORCEMENT LEARNING AND FEATURE DROPOUTBACKGROUND1 . Technical Field

[0001] This disclosure relates generally to improved machine learning techniques and, in non-limiting embodiments or aspects, to systems, methods, and computer program products for improved model training using reinforcement learning and feature dropout.2. Technical Considerations

[0002] Machine learning models are successful in performing many real-world tasks. However, performance of machine learning models can suffer for a variety of reasons that may be difficult to detect or understand. For example, overfitting may occur when a machine learning model is trained to fit too closely to training data. Additionally, unimportant and / or low-quality features of input data (e.g., training data and / or inference-time data) may bias the machine learning model and / or cause difficulty during training or inference (e.g., decrease training effectiveness, increase training time, reduce accuracy at inference time, and / or the like). Further, if there is a shift in the distribution of input data (e.g., at inference time compared to training / testing time, at a certain time period at inference time compared to other time periods, and / or the like), performance of the machine learning model may decrease if the model is not sufficiently robust.

[0003] Moreover, it is difficult to explain why a machine learning model generated a particular output and / or to interpret the behavior of a machine learning model. This difficulty is exacerbated (e.g., increases) as machine learning models become more complex and / or deeper (e.g., deep learning), and opacity of such models can be problematic. For example, in certain high-stakes scenarios such as financial investment and policy making, interpreting the behaviors of machine learning models is vital for humans to properly understand the outputs of the models prior to making important decisions. However, when interpretability is difficult, it can prevent understanding of why a model is performing poorly or why a model that previously performed well later starts to have performance degrade.

[0004] Certain techniques that attempt to address some of the deficiencies identified above are insufficient. For example, manually selecting features can be difficult, time consuming, and require a high level of understanding of the machine5Y77189.DOCX Page 1 of 69Attorney Docket No. 08223-2405237 (8270W001 ) learning model and the task being performed by the machine learning model. Separately, node dropout (e.g., randomly dropping out nodes of hidden layers, with each node having equal probability of being dropped) is inefficient (e.g., require a large amount of experimentation to tune the dropout rate as a hyperparameter) and lacks interpretability. Also, such node dropout would need to be different for different types of machine learning models (e.g., dropping out nodes of hidden layers of a deep neural network would not be directly achievable with a different type of machine learning model that does not include neural network and / or hidden layer structure).SUMMARY

[0005] Accordingly, provided are improved systems, methods, and computer program products for improved model training using reinforcement learning and feature dropout (e.g., that overcome some or all of the deficiencies identified above).

[0006] According to non-limiting embodiments or aspects, provided are systems for improved model training using reinforcement learning and feature dropout. An example system may include at least one processor configured to receive a pretrained machine learning model including a first input layer and a first plurality of hidden layers. The pretrained machine learning model may have been trained based on training data including a plurality of data records. Each data record of the plurality of data records may include a plurality of features. A second machine learning model may be initialized and may include a second input layer, a feature selection layer, and a second plurality of hidden layers. The second input layer may be initialized based on the first input layer, and the second plurality of hidden layers may be initialized based on the first plurality of hidden layers. The feature selection layer may include a plurality of weights comprising a respective weight for each respective feature of the plurality of features. Each respective weight of the plurality of weights may be associated with a respective dropout rate for the respective feature of the plurality of features. The plurality of weights of the feature selection layer of the second machine learning model may be adjusted based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.

[0007] In some non-limiting embodiments or aspects, initializing may include at least one of: copying parameters of the first input layer as the second input layer, copying parameters of the first plurality of hidden layers as the second plurality of hidden layers, or any combination thereof.5Y77189.DOCX Page 2 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0008] In some non-limiting embodiments or aspects, the at least one reinforcement learning reward may include at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0009] In some non-limiting embodiments or aspects, the at least one reinforcement learning reward may include the feature importance reward, the lossbased reward, and the divergence-based reward.

[0010] In some non-limiting embodiments or aspects, the feature importance reward may be based on a Pearson rank correlation and a reference feature importance ranking, the loss-based reward may be based on a cross-entropy loss, and / or the divergence-based reward may be based on a Kullback-Leibler (KL) divergence.

[0011] In some non-limiting embodiments or aspects, the feature importance reward may reward a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, the loss-based reward may penalize the reinforcement learning agent for a loss, and / or the divergence-based reward may penalize the reinforcement learning agent for a divergence.

[0012] In some non-limiting embodiments or aspects, a total reward is determined based on the feature importance reward, the loss-based reward, and / or the divergence-based reward. Additionally, or alternatively, adjusting the plurality of weights may include adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0013] In some non-limiting embodiments or aspects, the at least one processor may be further configured to retrain the second plurality of hidden layers based on the training data and the feature selection layer having the adjusted plurality of weights to provide adjusted parameters of the second plurality of hidden layers. For example, a tuned machine learning model may include the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.

[0014] In some non-limiting embodiments or aspects, the at least one processor may be further configured to deploy the pretrained machine learning model in a production environment to generate outputs based on incoming data records and / or detect at least one anomaly in the production environment. For example, receiving the5Y77189.DOCX Page 3 of 69Attorney Docket No. 08223-2405237 (8270W001 ) pretrained machine learning model may include receiving the pretrained machine learning model in a development environment based on detecting the at least one anomaly in the production environment. Additionally, or alternatively, initializing the second machine learning model may include initializing the second machine learning model in the development environment based on receiving the pretrained machine learning model in the development environment. Additionally, or alternatively, adjusting the plurality of weights of the feature selection layer may include adjusting the plurality of weights of the feature selection layer based on receiving the pretrained machine learning model in the development environment.

[0015] In some non-limiting embodiments or aspects, the at least one anomaly may include at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.

[0016] In some non-limiting embodiments or aspects, the at least one processor may be further configured to deploy the second machine learning model in the production environment.

[0017] According to non-limiting embodiments or aspects, provided are methods for improved model training using reinforcement learning and feature dropout. An example method may include receiving a pretrained machine learning model including a first input layer and a first plurality of hidden layers. The pretrained machine learning model may have been trained based on training data including a plurality of data records. Each data record of the plurality of data records may include a plurality of features. A second machine learning model may be initialized and may include a second input layer, a feature selection layer, and a second plurality of hidden layers. The second input layer may be initialized based on the first input layer. The second plurality of hidden layers may be initialized based on the first plurality of hidden layers. The feature selection layer may include a plurality of weights including a respective weight for each respective feature of the plurality of features. Each respective weight of the plurality of weights may be associated with a respective dropout rate for the respective feature of the plurality of features. The plurality of weights of the feature selection layer of the second machine learning model may be adjusted based on the5Y77189.DOCX Page 4 of 69Attorney Docket No. 08223-2405237 (8270W001 ) training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.

[0018] In some non-limiting embodiments or aspects, initializing may include at least one of: copying parameters of the first input layer as the second input layer, copying parameters of the first plurality of hidden layers as the second plurality of hidden layers, or any combination thereof.

[0019] In some non-limiting embodiments or aspects, the at least one reinforcement learning reward may include at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0020] In some non-limiting embodiments or aspects, the feature importance reward may be based on a Pearson rank correlation and a reference feature importance ranking. Additionally, or alternatively, the loss-based reward may be based on a cross-entropy loss. Additionally, or alternatively, the divergence-based reward is based on a Kullback-Leibler (KL) divergence.

[0021] In some non-limiting embodiments or aspects, the feature importance reward may reward a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking. Additionally, or alternatively, the loss-based reward may penalize the reinforcement learning agent for a loss. The divergence-based reward may penalize the reinforcement learning agent for a divergence.

[0022] In some non-limiting embodiments or aspects, a total reward may be determined based on the feature importance reward, the loss-based reward, and the divergence-based reward. Additionally, or alternatively, adjusting the plurality of weights may include adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0023] In some non-limiting embodiments or aspects, the second plurality of hidden layers may be retrained based on the training data and the feature selection layer having the adjusted plurality of weights to provide adjusted parameters of the second plurality of hidden layers. For example, a tuned machine learning model may include the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.5Y77189.DOCX Page 5 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0024] In some non-limiting embodiments or aspects, the pretrained machine learning model may be deployed in a production environment to generate outputs based on incoming data records. At least one anomaly may be detected in the production environment. In some non-limiting embodiments or aspects, receiving the pretrained machine learning model may include receiving the pretrained machine learning model in a development environment based on detecting the at least one anomaly in the production environment. Additionally, or alternatively, initializing the second machine learning model may include initializing the second machine learning model in the development environment based on receiving the pretrained machine learning model in the development environment. Additionally, or alternatively, adjusting the plurality of weights of the feature selection layer may include adjusting the plurality of weights of the feature selection layer based on receiving the pretrained machine learning model in the development environment.

[0025] In some non-limiting embodiments or aspects, the second machine learning model may be deployed in the production environment.

[0026] According to non-limiting embodiments or aspects, provided are computer program products for improved model training using reinforcement learning and feature dropout. An example computer program product may include at least one non- transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to receive a pretrained machine learning model including a first input layer and a first plurality of hidden layers. The pretrained machine learning model may have been trained based on training data comprising a plurality of data records. Each data record of the plurality of data records may include a plurality of features. A second machine learning model including a second input layer, a feature selection layer, and a second plurality of hidden layers may be initialized. The second input layer may be initialized based on the first input layer. The second plurality of hidden layers may be initialized based on the first plurality of hidden layers. The feature selection layer may include a plurality of weights including a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features. The plurality of weights of the feature selection layer of the second machine learning model may be adjusted based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.5Y77189.DOCX Page 6 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0027] According to non-limiting embodiments or aspects, provided are systems for improved model training using reinforcement learning and feature dropout. An example system may include a production environment system configured to deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records. The pretrained machine learning model may include a first input layer and a first plurality of hidden layers. The pretrained machine learning model may have been trained based on training data including a plurality of data records. Each data record of the plurality of data records may include a plurality of features. The production environment system may be configured to detect at least one anomaly in the production environment. A development environment system may be configured to, based on the production environment system detecting the at least one anomaly in the production environment, receive the pretrained machine learning model in a development environment. The development environment system may be configured to initialize a second machine learning model, which may include a second input layer, a feature selection layer, and a second plurality of hidden layers. The second input layer may be initialized based on the first input layer. The second plurality of hidden layers may be initialized based on the first plurality of hidden layers. The feature selection layer may include a plurality of weights including a respective weight for each respective feature of the plurality of features. Each respective weight of the plurality of weights may be associated with a respective dropout rate for the respective feature of the plurality of features. The development environment system may be configured to adjust the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights. The development environment system may be configured to retrain the second plurality of hidden layers based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers. A tuned machine learning model may include the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters. The development environment system may be configured to communicate the tuned machine learning model to the production environment system. The production environment system may be further5Y77189.DOCX Page 7 of 69Attorney Docket No. 08223-2405237 (8270W001 ) configured to deploy the tuned machine learning model in the production environment to generate further outputs based on further incoming data records.

[0028] In some non-limiting embodiments or aspects, initializing may include at least one of copying parameters of the first input layer as the second input layer, copying parameters of the first plurality of hidden layers as the second plurality of hidden layers, or any combination thereof.

[0029] In some non-limiting embodiments or aspects, the at least one reinforcement learning reward may include at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0030] In some non-limiting embodiments or aspects, the at least one reinforcement learning reward may include the feature importance reward, the lossbased reward, and divergence-based reward.

[0031] In some non-limiting embodiments or aspects, the feature importance reward may be based on a Pearson rank correlation and a reference feature importance ranking. Additionally, or alternatively, the loss-based reward may be based on a cross-entropy loss. Additionally, or alternatively, and the divergence-based reward is based on a Kullback-Leibler (KL) divergence.

[0032] In some non-limiting embodiments or aspects, the feature importance reward may reward a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking. Additionally, or alternatively, the loss-based reward may penalize the reinforcement learning agent for a loss. Additionally, or alternatively, the divergence-based reward may penalize the reinforcement learning agent for a divergence.

[0033] In some non-limiting embodiments or aspects, a total reward may be determined based on the feature importance reward, the loss-based reward, and the divergence-based reward. Additionally, or alternatively, adjusting the plurality of weights may include adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0034] In some non-limiting embodiments or aspects, the at least one anomaly may include at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth5Y77189.DOCX Page 8 of 69Attorney Docket No. 08223-2405237 (8270W001 ) distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.

[0035] According to non-limiting embodiments or aspects, provided are methods for improved model training using reinforcement learning and feature dropout. An example method may include deploying a pretrained machine learning model in a production environment to generate outputs based on incoming data records. The pretrained machine learning model may include a first input layer and a first plurality of hidden layers. The pretrained machine learning model may have been trained based on training data including a plurality of data records, each data record of the plurality of data records comprising a plurality of features. At least one anomaly may be detected in the production environment. Based on detecting the at least one anomaly in the production environment, the pretrained machine learning model may be received in a development environment. A second machine learning model may be initialized and may include a second input layer, a feature selection layer, and a second plurality of hidden layers. The second input layer may be initialized based on the first input layer. The second plurality of hidden layers may be initialized based on the first plurality of hidden layers. The feature selection layer may include a plurality of weights including a respective weight for each respective feature of the plurality of features. Each respective weight of the plurality of weights may be associated with a respective dropout rate for the respective feature of the plurality of features. The plurality of weights of the feature selection layer of the second machine learning model may be adjusted based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights. The second plurality of hidden layers may be retrained based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers. A tuned machine learning model may include the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters. The tuned machine learning model may be deployed in the production environment to generate further outputs based on further incoming data records.5Y77189.DOCX Page 9 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0036] In some non-limiting embodiments or aspects, initializing may include at least one of copying parameters of the first input layer as the second input layer, copying parameters of the first plurality of hidden layers as the second plurality of hidden layers, or any combination thereof.

[0037] In some non-limiting embodiments or aspects, the at least one reinforcement learning reward may include at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0038] In some non-limiting embodiments or aspects, the feature importance reward may reward a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking. Additionally, or alternatively, the loss-based reward may penalize the reinforcement learning agent for a loss. The divergence-based reward may penalize the reinforcement learning agent for a divergence.

[0039] In some non-limiting embodiments or aspects, a total reward may be determined based on the feature importance reward, the loss-based reward, and divergence-based reward. Additionally, or alternatively, adjusting the plurality of weights may include adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0040] In some non-limiting embodiments or aspects, the at least one anomaly may include at least one of a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.

[0041] According to non-limiting embodiments or aspects, provided are computer program products for improved model training using reinforcement learning and feature dropout. An example computer program product may include at least one non- transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records. The pretrained machine learning model may include a first input layer and a first plurality of hidden layers. The pretrained machine learning5Y77189.DOCX Page 10 of 69Attorney Docket No. 08223-2405237 (8270W001 ) model may have been trained based on training data comprising a plurality of data records. Each data record of the plurality of data records may include a plurality of features. At least one anomaly may be detected in the production environment. Based on detecting the at least one anomaly in the production environment, the pretrained machine learning model may be received in a development environment. A second machine learning model may be initialized and may include a second input layer, a feature selection layer, and a second plurality of hidden layers. The second input layer may be initialized based on the first input layer. The second plurality of hidden layers may be initialized based on the first plurality of hidden layers. The feature selection layer may include a plurality of weights including a respective weight for each respective feature of the plurality of features. Each respective weight of the plurality of weights may be associated with a respective dropout rate for the respective feature of the plurality of features. The plurality of weights of the feature selection layer of the second machine learning model may be adjusted based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights. The second plurality of hidden layers may be retrained based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers. A tuned machine learning model may include the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters. The tuned machine learning model may be deployed in the production environment to generate further outputs based on further incoming data records.

[0042] According to non-limiting embodiments or aspects, provided are computer program products for improved model training using reinforcement learning and feature dropout. An example computer program product may include at least one non- transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to perform any of the methods described herein.

[0043] Further non-limiting embodiments or aspects are set forth in the following numbered clauses:5Y77189.DOCX Page 1 1 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0044] Clause 1 : A system, comprising: at least one processor configured to: receive a pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; and adjust the plurality of weights of the feature selection layer of the second machine learning model based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.

[0045] Clause 2: The system of clause 1 , wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

[0046] Clause 3: The system of clause 1 or clause 2, wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0047] Clause 4: The system of any of clauses 1 -3, wherein the at least one reinforcement learning reward comprises the feature importance reward, the lossbased reward, and the divergence-based reward.

[0048] Clause 5: The system of any of clauses 1 -4, wherein the feature importance reward is based on a Pearson rank correlation and a reference feature importance ranking, the loss-based reward is based on a cross-entropy loss, and the divergencebased reward is based on a Kullback-Leibler (KL) divergence.

[0049] Clause 6: The system of any of clauses 1 -5, wherein the feature importance reward rewards a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement5Y77189.DOCX Page 12 of 69Attorney Docket No. 08223-2405237 (8270W001 ) learning agent for a loss, and wherein the divergence-based reward penalizes the reinforcement learning agent for a divergence.

[0050] Clause 7: The system of any of clauses 1 -6, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and the divergence-based reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0051] Clause 8: The system of any of clauses 1 -7, wherein the at least one processor is further configured to: retrain the second plurality of hidden layers based on the training data and the feature selection layer having the adjusted plurality of weights to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.

[0052] Clause 9: The system of any of clauses 1 -8, wherein the at least one processor is further configured to: deploy the pretrained machine learning model in a production environment to generate outputs based on incoming data records; and detect at least one anomaly in the production environment; wherein receiving the pretrained machine learning model comprises receiving the pretrained machine learning model in a development environment based on detecting the at least one anomaly in the production environment, wherein initializing the second machine learning model comprises initializing the second machine learning model in the development environment based on receiving the pretrained machine learning model in the development environment, and wherein adjusting the plurality of weights of the feature selection layer comprises adjusting the plurality of weights of the feature selection layer based on receiving the pretrained machine learning model in the development environment.

[0053] Clause 10: The system of any of clauses 1 -9, wherein the at least one anomaly comprises at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase5Y77189.DOCX Page 13 of 69Attorney Docket No. 08223-2405237 (8270W001 ) in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.

[0054] Clause 1 1 : The system of any of clauses 1 -10, wherein the at least one processor is further configured to: deploy the second machine learning model in the production environment.

[0055] Clause 12: A computer-implemented method, comprising: receiving, with at least one processor, a pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; initializing, with at least one processor, a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; and adjusting, with at least one processor, the plurality of weights of the feature selection layer of the second machine learning model based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.

[0056] Clause 13: The computer-implemented method of clause 12, wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

[0057] Clause 14: The computer-implemented method of clause 12 or clause 13, wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0058] Clause 15: The computer-implemented method of any of clauses 12-14, wherein the feature importance reward is based on a Pearson rank correlation and a reference feature importance ranking, the loss-based reward is based on a crossentropy loss; and the divergence-based reward is based on a Kullback-Leibler (KL) divergence.

[0059] Clause 16: The computer-implemented method of any of clauses 12-15, wherein the feature importance reward rewards a reinforcement learning agent when5Y77189.DOCX Page 14 of 69Attorney Docket No. 08223-2405237 (8270W001 ) a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement learning agent for a loss, and wherein the divergencebased reward penalizes the reinforcement learning agent for a divergence, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and the divergence-based reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0060] Clause 17: The computer-implemented method of any of clauses 12-16, further comprising retraining, with at least one processor, the second plurality of hidden layers based on the training data and the feature selection layer having the adjusted plurality of weights to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.

[0061] Clause 18: The computer-implemented method of any of clauses 12-17, further comprising: deploying, with at least one processor, the pretrained machine learning model in a production environment to generate outputs based on incoming data records; and detecting, with at least one processor, at least one anomaly in the production environment; wherein receiving the pretrained machine learning model comprises receiving the pretrained machine learning model in a development environment based on detecting the at least one anomaly in the production environment, wherein initializing the second machine learning model comprises initializing the second machine learning model in the development environment based on receiving the pretrained machine learning model in the development environment, and wherein adjusting the plurality of weights of the feature selection layer comprises adjusting the plurality of weights of the feature selection layer based on receiving the pretrained machine learning model in the development environment.

[0062] Clause 19: The computer-implemented method of any of clauses 12-18, further comprising deploying, with at least one processor, the second machine learning model in the production environment.5Y77189.DOCX Page 15 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0063] Clause 20: A computer program product comprising at least one non- transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to: receive a pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; and adjust the plurality of weights of the feature selection layer of the second machine learning model based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.

[0064] Clause 21 : A system, comprising: a production environment system configured to: deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records, the pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; and detect at least one anomaly in the production environment; and a development environment system configured to: based on the production environment system detecting the at least one anomaly in the production environment, receive the pretrained machine learning model in a development environment; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; adjust the plurality of weights of the feature selection layer of the second5Y77189.DOCX Page 16 of 69Attorney Docket No. 08223-2405237 (8270W001 ) machine learning model based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights; retrain the second plurality of hidden layers based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters; and communicate the tuned machine learning model to the production environment system, wherein the production environment system is further configured to: deploy the tuned machine learning model in the production environment to generate further outputs based on further incoming data records.

[0065] Clause 22: The system of clause 21 , wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

[0066] Clause 23: The system of clause 21 or clause 22, wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0067] Clause 24: The system of any of clauses 21 -23, wherein the at least one reinforcement learning reward comprises the feature importance reward, the lossbased reward, and divergence-based reward.

[0068] Clause 25: The system of any of clauses 21 -24, wherein the feature importance reward is based on a Pearson rank correlation and a reference feature importance ranking, the loss-based reward is based on a cross-entropy loss; and the divergence-based reward is based on a Kullback-Leibler (KL) divergence.

[0069] Clause 26: The system of any of clauses 21 -25, wherein the feature importance reward rewards a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement learning agent for a loss, and wherein the divergence-based reward penalizes the reinforcement learning agent for a divergence.5Y77189.DOCX Page 17 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0070] Clause 27: The system of any of clauses 21 -26, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and the divergence-based reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0071] Clause 28: The system of any of clauses 21 -27, wherein the at least one anomaly comprises at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.

[0072] Clause 29: A computer-implemented method, comprising: deploying, with at least one processor, a pretrained machine learning model in a production environment to generate outputs based on incoming data records, the pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; and detecting, with at least one processor, at least one anomaly in the production environment; based on detecting the at least one anomaly in the production environment: receiving, with at least one processor, the pretrained machine learning model in a development environment; initializing, with at least one processor, a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; adjusting, with at least one processor, the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights; retraining, with at least one processor, the second plurality of hidden layers based on5Y77189.DOCX Page 18 of 69Attorney Docket No. 08223-2405237 (8270W001 ) the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters; and deploying, with at least one processor, the tuned machine learning model in the production environment to generate further outputs based on further incoming data records.

[0073] Clause 30: The computer-implemented method of clause 29, wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

[0074] Clause 31 : The computer-implemented method of clause 29 or clause 30, wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

[0075] Clause 32: The computer-implemented method of any of clause 29-31 , wherein the feature importance reward rewards a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement learning agent for a loss, and wherein the divergencebased reward penalizes the reinforcement learning agent for a divergence.

[0076] Clause 33: The computer-implemented method of any of clause 29-32, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and divergence-based reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

[0077] Clause 34: The computer-implemented method of any of clause 29-33, wherein the at least one anomaly comprises at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the5Y77189.DOCX Page 19 of 69Attorney Docket No. 08223-2405237 (8270W001 ) pretrained machine learning model based on the incoming data records, or any combination thereof.

[0078] Clause 35: A computer program product comprising at least one non- transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to: deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records, the pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; and detect at least one anomaly in the production environment; based on detecting the at least one anomaly in the production environment: receive the pretrained machine learning model in a development environment; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; adjust the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights; retrain the second plurality of hidden layers based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters; and deploy the tuned machine learning model in the production environment to generate further outputs based on further incoming data records.

[0079] Clause 36: A computer program product comprising at least one non- transitory computer-readable medium including program instructions that, when5Y77189.DOCX Page 20 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) executed by at least one processor, cause the at least one processor to perform the method of any of clauses 12-19 or 29-34.

[0080] These and other features and characteristics of the present disclosure, as well as the methods of operation and functions of the related elements of structures and the combination of parts and economies of manufacture, will become more apparent upon consideration of the following description and the appended claims with reference to the accompanying drawings, all of which form a part of this specification, wherein like reference numerals designate corresponding parts in the various figures. It is to be expressly understood, however, that the drawings are for the purpose of illustration and description only and are not intended as a definition of the limits of the disclosed subject matter.BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Additional advantages and details are explained in greater detail below with reference to the non-limiting, exemplary embodiments that are illustrated in the accompanying schematic figures, in which:

[0082] FIG. 1 is a schematic diagram of a system for improved model training using reinforcement learning and feature dropout, according to some non-limiting embodiments or aspects;

[0083] FIG. 2A is a flow diagram of a method for improved model training using reinforcement learning and feature dropout, according to some non-limiting embodiments or aspects;

[0084] FIG. 2B is a flow diagram of a method for improved model training using reinforcement learning and feature dropout, according to some non-limiting embodiments or aspects;

[0085] FIG. 3 is a diagram of an example payment processing network in which methods, systems, and / or computer program products, described herein, may be implemented, according to some non-limiting embodiments or aspects;

[0086] FIG. 4 is a schematic diagram of example components of one or more devices of FIG. 1 and / or FIG. 3, according to some non-limiting embodiments or aspects;

[0087] FIGS. 5A-5C are schematic diagrams of an example implementation of systems and methods for improved model training using reinforcement learning and feature dropout, according to some non-limiting embodiments or aspects; and5Y77189.DOCX Page 21 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0088] FIG. 6 is a graph showing performance of the disclosed techniques for improved model training using reinforcement learning and feature dropout compared to other techniques for model training, according to some non-limiting embodiments or aspects.DETAILED DESCRIPTION

[0089] For purposes of the description hereinafter, the terms “end,” “upper,” “lower,” “right,” “left,” “vertical,” “horizontal,” “top,” “bottom,” “lateral,” “longitudinal,” and derivatives thereof shall relate to the embodiments as they are oriented in the drawing figures. However, it is to be understood that the present disclosure may assume various alternative variations and step sequences, except where expressly specified to the contrary. It is also to be understood that the specific devices and processes illustrated in the attached drawings, and described in the following specification, are simply exemplary and non-limiting embodiments or aspects of the disclosed subject matter. Hence, specific dimensions and other physical characteristics related to the embodiments or aspects disclosed herein are not to be considered as limiting.

[0090] Some non-limiting embodiments or aspects are described herein in connection with thresholds. As used herein, satisfying a threshold may refer to a value being greater than the threshold, more than the threshold, higher than the threshold, greater than or equal to the threshold, less than the threshold, fewer than the threshold, lower than the threshold, less than or equal to the threshold, equal to the threshold, etc.

[0091] No aspect, component, element, structure, act, step, function, instruction, and / or the like used herein should be construed as critical or essential unless explicitly described as such. Also, as used herein, the articles “a” and “an” are intended to include one or more items and may be used interchangeably with “one or more” and “at least one.” Furthermore, as used herein, the term “set” is intended to include one or more items (e.g., related items, unrelated items, a combination of related and unrelated items, and / or the like) and may be used interchangeably with “one or more” or “at least one.” Where only one item is intended, the term “one” or similar language is used. Also, as used herein, the terms “has,” “have,” “having,” or the like are intended to be open-ended terms. Further, the phrase “based on” is intended to mean “based at least partially on” unless explicitly stated otherwise. In addition, reference to an action being “based on” a condition may refer to the action being “in response to” the condition. For example, the phrases “based on” and “in response to” may, in some5Y77189.DOCX Page 22 of 69Attorney Docket No. 08223-2405237 (8270W001 ) non-limiting embodiments or aspects, refer to a condition for automatically triggering an action (e.g., a specific operation of an electronic device, such as a computing device, a processor, and / or the like).

[0092] As used herein, the term “acquirer institution” may refer to an entity licensed and / or approved by a transaction service provider to originate transactions (e.g., payment transactions) using a payment device associated with the transaction service provider. The transactions the acquirer institution may originate may include payment transactions (e.g., purchases, original credit transactions (OCTs), account funding transactions (AFTs), and / or the like). In some non-limiting embodiments or aspects, an acquirer institution may be a financial institution, such as a bank. As used herein, the term “acquirer system” may refer to one or more computing devices operated by or on behalf of an acquirer institution, such as a server computer executing one or more software applications.

[0093] As used herein, the term “account identifier” may include one or more primary account numbers (PANs), tokens, or other identifiers associated with a customer account. The term “token” may refer to an identifier that is used as a substitute or replacement identifier for an original account identifier, such as a PAN. Account identifiers may be alphanumeric or any combination of characters and / or symbols. Tokens may be associated with a PAN or other original account identifier in one or more data structures (e.g., one or more databases, and / or the like) such that they may be used to conduct a transaction without directly using the original account identifier. In some examples, an original account identifier, such as a PAN, may be associated with a plurality of tokens for different individuals or purposes.

[0094] As used herein, the terms “client” and “client device” may refer to one or more client-side devices or systems (e.g., remote from a transaction service provider) used to initiate or facilitate a transaction (e.g., a payment transaction). As an example, a “client device” may refer to one or more POS devices used by a merchant, one or more acquirer host computers used by an acquirer, one or more mobile devices used by a user, and / or the like. In some non-limiting embodiments or aspects, a client device may be an electronic device configured to communicate with one or more networks and initiate or facilitate transactions. For example, a client device may include one or more computers, portable computers, laptop computers, tablet computers, mobile devices, cellular phones, wearable devices (e.g., watches, glasses, lenses, clothing, and / or the like), PDAs, and / or the like. Moreover, a “client” may also refer to an entity5Y77189.DOCX Page 23 of 69Attorney Docket No. 08223-2405237 (8270W001 )(e.g., a merchant, an acquirer, and / or the like) that owns, utilizes, and / or operates a client device for initiating transactions (e.g., for initiating transactions with a transaction service provider).

[0095] As used herein, the term “communication” may refer to the reception, receipt, transmission, transfer, provision, and / or the like of data (e.g., information, signals, messages, instructions, commands, and / or the like). For one unit (e.g., a device, a system, a component of a device or system, combinations thereof, and / or the like) to be in communication with another unit means that the one unit is able to directly or indirectly receive information from and / or transmit information to the other unit. This may refer to a direct or indirect connection (e.g., a direct communication connection, an indirect communication connection, and / or the like) that is wired and / or wireless in nature. Additionally, two units may be in communication with each other even though the information transmitted may be modified, processed, relayed, and / or routed between the first and second unit. For example, a first unit may be in communication with a second unit even though the first unit passively receives information and does not actively transmit information to the second unit. As another example, a first unit may be in communication with a second unit if at least one intermediary unit processes information received from the first unit and communicates the processed information to the second unit. In some non-limiting embodiments or aspects, a message may refer to a network packet (e.g., a data packet and / or the like) that includes data. It will be appreciated that numerous other arrangements are possible.

[0096] As used herein, the term “computing device” may refer to one or more electronic devices configured to process data. A computing device may, in some examples, include the necessary components to receive, process, and output data, such as a processor, a display, a memory, an input device, a network interface, and / or the like. A computing device may be a mobile device. As an example, a mobile device may include a cellular phone (e.g., a smartphone or standard cellular phone), a portable computer, a wearable device (e.g., watches, glasses, lenses, clothing, and / or the like), a personal digital assistant (PDA), and / or other like devices. A computing device may also be a desktop computer or other form of non-mobile computer.

[0097] As used herein, the terms “electronic wallet” and “electronic wallet application” refer to one or more electronic devices and / or software applications configured to initiate and / or conduct payment transactions. For example, an electronic5Y77189.DOCX Page 24 of 69Attorney Docket No. 08223-2405237 (8270W001 ) wallet may include a mobile device executing an electronic wallet application, and may further include server-side software and / or databases for maintaining and providing transaction data to the mobile device. An “electronic wallet provider” may include an entity that provides and / or maintains an electronic wallet for a customer, such as Google Pay®, Android Pay®, Apple Pay®, Samsung Pay®, and / or other like electronic payment systems. In some non-limiting examples, an issuer bank may be an electronic wallet provider.

[0098] As used herein, the term “issuer institution” may refer to one or more entities, such as a bank, that provide accounts to customers for conducting transactions (e.g., payment transactions), such as initiating credit and / or debit payments. For example, an issuer institution may provide an account identifier, such as a PAN, to a customer that uniquely identifies one or more accounts associated with that customer. The account identifier may be embodied on a portable financial device, such as a physical financial instrument, e.g., a payment card, and / or may be electronic and used for electronic payments. The term “issuer system” refers to one or more computer devices operated by or on behalf of an issuer institution, such as a server computer executing one or more software applications. For example, an issuer system may include one or more authorization servers for authorizing a transaction.

[0099] As used herein, the term “merchant” may refer to an individual or entity that provides goods and / or services, or access to goods and / or services, to customers based on a transaction, such as a payment transaction. The term “merchant” or “merchant system” may also refer to one or more computer systems operated by or on behalf of a merchant, such as a server computer executing one or more software applications.

[0100] As used herein, a “point-of-sale (POS) device” may refer to one or more devices, which may be used by a merchant to conduct a transaction (e.g., a payment transaction) and / or process a transaction. For example, a POS device may include one or more client devices. Additionally, or alternatively, a POS device may include peripheral devices, card readers, scanning devices (e.g., code scanners), Bluetooth® communication receivers, near-field communication (NFC) receivers, radio frequency identification (RFID) receivers, and / or other contactless transceivers or receivers, contact-based receivers, payment terminals, and / or the like. As used herein, a “point- of-sale (POS) system” may refer to one or more client devices and / or peripheral devices used by a merchant to conduct a transaction. For example, a POS system5Y77189.DOCX Page 25 of 69Attorney Docket No. 08223-2405237 (8270W001 ) may include one or more POS devices and / or other like devices that may be used to conduct a payment transaction. In some non-limiting embodiments or aspects, a POS system (e.g., a merchant POS system) may include one or more server computers programmed or configured to process online payment transactions through webpages, mobile applications, and / or the like.

[0101] As used herein, the term “payment device” may refer to a payment card (e.g., a credit or debit card), a gift card, a smartcard, smart media, a payroll card, a healthcare card, a wristband, a machine-readable medium containing account information, a keychain device or fob, an RFID transponder, a retailer discount or loyalty card, a cellular phone, an electronic wallet mobile application, a personal digital assistant (PDA), a pager, a security card, a computing device, an access card, a wireless terminal, a transponder, and / or the like. In some non-limiting embodiments or aspects, the payment device may include volatile or non-volatile memory to store information (e.g., an account identifier, a name of the account holder, and / or the like).

[0102] As used herein, the term “payment gateway” may refer to an entity and / or a payment processing system operated by or on behalf of such an entity (e.g., a merchant service provider, a payment service provider, a payment facilitator, a payment facilitator that contracts with an acquirer, a payment aggregator, and / or the like), which provides payment services (e.g., transaction service provider payment services, payment processing services, and / or the like) to one or more merchants. The payment services may be associated with the use of portable financial devices managed by a transaction service provider. As used herein, the term “payment gateway system” may refer to one or more computer systems, computer devices, servers, groups of servers, and / or the like, operated by or on behalf of a payment gateway.

[0103] As used herein, the term “server” may refer to or include one or more computing devices that are operated by or facilitate communication and processing for multiple parties in a network environment, such as the Internet, although it will be appreciated that communication may be facilitated over one or more public or private network environments and that various other arrangements are possible. Further, multiple computing devices (e.g., servers, point-of-sale (POS) devices, mobile devices, etc.) directly or indirectly communicating in the network environment may constitute a “system.”5Y77189.DOCX Page 26 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0104] As used herein, the term “system” may refer to one or more computing devices or combinations of computing devices (e.g., processors, servers, client devices, software applications, components of such, and / or the like). Reference to “a device,” “a server,” “a processor,” and / or the like, as used herein, may refer to a previously-recited device, server, or processor that is recited as performing a previous step or function, a different device, server, or processor, and / or a combination of devices, servers, and / or processors. For example, as used in the specification and the claims, a first device, a first server, or a first processor that is recited as performing a first step or a first function may refer to the same or different device, server, or processor recited as performing a second step or a second function.

[0105] As used herein, the term “transaction service provider” may refer to an entity that receives transaction authorization requests from merchants or other entities and provides guarantees of payment, in some cases through an agreement between the transaction service provider and an issuer institution. For example, a transaction service provider may include a payment network such as Visa® or any other entity that processes transactions. The term “transaction processing system” may refer to one or more computer systems operated by or on behalf of a transaction service provider, such as a transaction processing server executing one or more software applications. A transaction processing server may include one or more processors and, in some non-limiting embodiments or aspects, may be operated by or on behalf of a transaction service provider.

[0106] Non-limiting embodiments or aspects of the disclosed subject matter are directed to systems, methods, and computer program products for improved machine learning model training, including, but not limited to, improved machine learning model training using reinforcement learning and feature dropout. For example, non-limiting embodiments or aspects of the disclosed subject matter provide receiving a pretrained machine learning model (e.g., having an input layer and hidden layers) that was trained based on training data having a plurality of features and initializing a second machine learning model including a feature selection layer (e.g., in addition to an input layer and hidden layers). For example, the input layer and hidden layers of the second machine learning model may be initialized based on (e.g., copy the parameters of) the input layer and hidden layers of the pretrained machine learning model. The feature selection layer may include a plurality of weights including a respective weight for each respective feature, and each respective weight may be associated with a respective5Y77189.DOCX Page 27 of 69Attorney Docket No. 08223-2405237 (8270W001 ) dropout rate for the respective feature. The weights of the feature selection layer may be adjusted based on the training data and at least one (e.g., a plurality of) reinforcement learning reward(s) to provide an adjusted plurality of weights. In this way, robustness and performance of the machine learning model may be improved by automatically removing or decreasing the impact of features with poor quality using the trained feature selection layer. For example, the automatic features selection and flexible dropout rate provided by the feature selection layer during training (or retraining) of the machine learning model reduce overfitting and reduce performance drop caused by poor data quality of a particular feature. Also, the feature selection resulting from the feature selection layer does not require domain knowledge of the input data features, the model, or the task(s) being performed. Additionally, interpretability of the machine learning model is improved because the importance of features can be inferred from the weights of the trained feature selection layer (e.g., a higher dropout rate may be less important, a lower dropout rate may be more important, and / or the like). Moreover, efficiency is improved by automating feature importance analysis and hyperparameter tuning (e.g., tuning dropout rate) using reinforcement learning, both of which would otherwise require manual efforts and / or many inefficient iterations. Furthermore, the disclosed techniques are flexible and can be applied to any deep learning model (e.g., because the feature selection layer may be applied at the input layer regardless of the model structure after the input layer). The reinforcement learning rewards can also be customized to the machine learning model being tuned. For example, one of the reinforcement learning rewards can be based on the same loss function (or same type of loss function) used to pretrain the machine learning model.

[0107] Further, non-limiting embodiments or aspects of the disclosed subject matter provide deploying a pretrained machine learning model in a production environment to generate outputs based on incoming data records. At least one anomaly may be detected in the production environment. Based on detecting the anomaly in the production environment, the pretrained machine learning model may be received in a development environment, where a second machine learning model including a feature selection layer may be initialized, as described herein. The weights of the feature selection layer of the second machine learning model may be adjusted based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, any combination thereof, and / or the like. The hidden5Y77189.DOCX Page 28 of 69Attorney Docket No. 08223-2405237 (8270W001 ) layers of the second machine learning model may be retrained based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, any combination thereof, and / or the like to provide adjusted parameters of the hidden layers. Optionally, the reinforcement learning for the feature selection layer and the retraining of the hidden layers may be repeated. The result of the reinforcement learning for the feature selection layer and the retraining of the hidden layers may be a tuned machine learning model, which may be deployed in the production environment (e.g., in place of the original pretrained model) to generate further outputs based on further incoming data records. In addition to the improvements and advantages described above, such techniques allow for automatic tuning / retraining of a machine learning model to replace a pretrained model (e.g., after an anomaly, such as at least one of a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a first distribution of the incoming data records for a first time period with respect to a second distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, any combination thereof, and / or the like). As such, reduced performance resulting from the anomaly may be reduced or eliminated by replacing the pretrained machine learning model with the tuned machine learning model. Such automatic, efficient, and swift replacement may be especially beneficial for high stakes scenarios in which the result of real-world decisions made based on the outputs of the machine learning model can have serious consequences.

[0108] For example, financial transactions, financial investment, policy making, cyber security, medical treatment, and / or the like are examples of such high stakes scenarios. For the purpose of illustration, the machine learning model may be used to predict clearing and / or settlement of previously authorized payment transactions, and financial institutions may post transactions to accounts and / or shift large amounts of money based on those predictions. As another example, the machine learning model may be used to predict whether a payment transaction is fraudulent, and the payment transaction may be authorized or denied based on the outputs of the machine learning model. The machine learning model may be used to detect anomalies, and certain remediation actions may be taken or not taken based on the outputs of the machine learning model. Similarly, the machine learning model may be used for authentication and / or identity verification, and access (e.g., to sensitive information, secure physical5Y77189.DOCX Page 29 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) locations, and / or the like) may be controlled based on the outputs of the machine learning model. Further still, the machine learning model may be a large language model (LLM) used to predict a certain type of risk (e.g., financial risk, personal safety risk, and / or the like) based on reports (e.g., news reports, social media posts, and / or the like), and real-world decisions may be made based on the outputs of the machine learning model.

[0109] Referring now to FIG. 1 , shown is an example system 100 for improved model training using reinforcement learning and feature dropout, according to some non-limiting embodiments or aspects. As shown in FIG. 1 , system 100 may include development environment system 102, data storage system 104, production environment system 106, and / or communication network 1 12.

[0110] Development environment system 102 may be used for developing (e.g., generating, training, testing, tuning, retraining, and / or the like) machine learning models. Development environment system 102 may include one or more devices capable of receiving information from and / or communicating information to data storage system 104 and / or production environment system 106 (e.g., directly via wired or wireless communication connection, indirectly via communication network 112, and / or the like). For example, development environment system 102 may include a computing device, such as a server, a group of servers, a desktop computer, a portable computer, a mobile device, and / or other like devices. In some non-limiting embodiments or aspects, development environment system 102 may be in communication with a data storage device (e.g., data storage system 104), which may be local or remote to development environment system 102. In some non-limiting embodiments or aspects, development environment system 102 may be capable of receiving information from, storing information in, communicating information to, or searching information stored in the data storage device (e.g., data storage system 104).

[0111] Data storage system 104 may include one or more devices capable of receiving information from and / or communicating information to development environment system 102 and / or production environment system 106 (e.g., directly via wired or wireless communication connection, indirectly via communication network 1 12, and / or the like). For example, data storage system 104 may include a computing device, such as a server, a group of servers, and / or other like devices. In some nonlimiting embodiments or aspects, data storage system 104 may include a data storage5Y77189.DOCX Page 30 of 69Attorney Docket No. 08223-2405237 (8270W001 ) device. In some non-limiting embodiments or aspects, data storage system 104 may be capable of receiving information from, storing information in, communicating information to, or searching information stored in the data storage device. In some non-limiting embodiments or aspects, data storage system 104 may be part of development environment system 102 and / or part of the same system as development environment system 102.

[0112] Production environment system 106 may be used for deploying (e.g., making available to end users, making available to real-world applications, making available to generate real-time or near real-time outputs (e.g., predictions, inferences, and / or the like) based on incoming (e.g., real-world) data, and / or the like) machine learning models (e.g., trained, pretrained, tuned, and / or the like machine learning models). Production environment system 106 may include one or more devices capable of receiving information from and / or communicating information to development environment system 102 and / or data storage system 104 (e.g., directly via wired or wireless communication connection, indirectly via communication network 1 12, and / or the like). For example, production environment system 106 may include a computing device, such as a server, a group of servers, a desktop computer, a portable computer, a mobile device, and / or other like devices. In some non-limiting embodiments or aspects, production environment system 106 may be part of the same system as development environment system 102. For example, development environment system 102 and production environment system 106 may be subsystems of a larger system, and / or development environment system 102 and production environment system 106 may be logically or physically separate.

[0113] In some non-limiting embodiments or aspects, production environment system 106 may deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records, as described herein. For example, the pretrained machine learning model may include a first input layer and a first plurality of hidden layers. The pretrained machine learning model may have been trained based on training data comprising a plurality of data records. Each data record may include a plurality of features.

[0114] In some non-limiting embodiments or aspects, production environment system 106 may detect at least one anomaly in the production environment, as described herein. For example, production environment system 106 may detect at least one of a shift between a first distribution of the incoming data records with respect5Y77189.DOCX Page 31 of 69Attorney Docket No. 08223-2405237 (8270W001 ) to a second distribution of the training data, a shift between a first distribution of the incoming data records for a first time period with respect to a second distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, any combination thereof, and / or the like.

[0115] In some non-limiting embodiments or aspects, development environment system 102 may receive a pretrained machine learning model, as described herein. For example, development environment system 102 may receive the pretrained machine learning model in a development environment based on production environment system 106 detecting the at least one anomaly in the production environment. Additionally, or alternatively, development environment system 102 may receive the pretrained machine learning model at any time after pretraining of the machine learning model is complete. In some non-limiting embodiments or aspects, receiving the pretrained machine learning model may include development environment system 102 retrieving the pretrained machine learning model from data storage system 104. For example, development environment system 102 may retrieve the pretrained machine learning model from data storage system 104 based on production environment system 106 detecting the at least one anomaly.

[0116] In some non-limiting embodiments or aspects, development environment system 102 may initialize a second machine learning model, as described herein. For example, the second machine learning model may include a second input layer, a feature selection layer, and a second plurality of hidden layers. In some non-limiting embodiments or aspects, the input layer and the hidden layers of the second machine learning model may be initialized based on (e.g., copy the parameters from) the input layer and the hidden layers of the pretrained machine learning model. For example, the second machine learning model may have a structure that is the same as the pretrained machine learning model except for the addition of a feature selection layer (e.g., between the input layer and the hidden layer(s)), and the parameters of the second machine learning model may initially be the same as the pretrained machine learning model (except for the feature selection layer). In some non-limiting embodiments or aspects, the feature selection layer may include a plurality of weights including a respective weight for each respective feature of the plurality of features. For example, each respective weight of the plurality of weights may be associated with a respective dropout rate for the respective feature of the plurality of features.5Y77189.DOCX Page 32 of 69Attorney Docket No. 08223-2405237 (8270WG01 )

[0117] In some non-limiting embodiments or aspects, development environment system 102 may adjust the weights of the feature selection layer of the second machine learning model based on reinforcement learning. For example, development environment system 102 may adjust the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of the training data, the incoming data records, any combination thereof, and / or the like to provide an adjusted plurality of weights.

[0118] In some non-limiting embodiments or aspects, development environment system 102 may retrain the hidden layers of the second machine learning model, as described herein. For example, development environment system 102 may retrain the hidden layers of the second machine learning model based on the feature selection layer having the adjusted plurality of weights and at least one of the training data, the incoming data records, any combination thereof, and / or the like to provide adjusted parameters of the second plurality of hidden layers. For example, a tuned machine learning model may include the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.

[0119] In some non-limiting embodiments or aspects, development environment system 102 may communicate with production environment system 106 based on the tuned machine learning model. For example, development environment system 102 may communicate the tuned machine learning model to production environment system 106 (and / or production environment system 106 may retrieve the tuned machine learning model from data storage system 104 based on the communication from development environment system 102).

[0120] In some non-limiting embodiments or aspects, production environment system 106 may deploy the tuned machine learning model in the production environment to generate further outputs based on further incoming data records, as described herein.

[0121] The number and arrangement of systems and devices shown in FIG. 1 are provided as an example. There may be additional systems and / or devices, fewer systems and / or devices, different systems and / or devices, and / or differently arranged systems and / or devices than those shown in FIG. 1. Furthermore, two or more systems or devices shown in FIG. 1 may be implemented within a single system or device, or a single system or device shown in FIG. 1 may be implemented as multiple,5Y77189.DOCX Page 33 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) distributed systems or devices. Additionally, or alternatively, a set of systems (e.g., one or more systems) or a set of devices (e.g., one or more devices) of system 100 may perform one or more functions described as being performed by another set of systems or another set of devices of system 100.

[0122] Referring now to FIG. 2A, shown is a flow diagram for an example method for 200a, according to some non-limiting embodiments or aspects. The steps shown in FIG. 2A are for example purposes only. It will be appreciated that additional, fewer, different, and / or a different order of steps may be used in some non-limiting embodiments or aspects. In some non-limiting embodiments or aspects, a step may be automatically performed in response to performance and / or completion of a prior step. In some non-limiting embodiments or aspects, one or more of the steps of method 200a may be performed (e.g., completely, partially, and / or the like) by development environment system 102 (e.g., one or more devices of development environment system 102). In some non-limiting embodiments or aspects, one or more of the steps of method 200a may be performed (e.g., completely, partially, and / or the like) by another system, another device, another group of systems, or another group of devices, separate from or including development environment system 102, such as data storage system 104, production environment system 106, and / or the like.

[0123] As shown in FIG. 2A, at step 212, method 200a may include receiving a first (e.g., pretrained) machine learning model. For example, development environment system 102 may receive a pretrained machine learning model, as described herein.

[0124] In some non-limiting embodiments or aspects, the pretrained machine learning model may include a first input layer and a first plurality of hidden layers. In some non-limiting embodiments or aspects, the pretrained machine learning model may include at least one of a neural network, a convolutional neural network (CNN), a recurrent neural network (RNN), an autoencoder (or portion thereof), a long short-term memory (LSTM), a bidirectional LSTM, an attention network, a transformer, a treebased model, a decision tree model, a random forest model, a gradient boosting model, an XGBoost model, any combination thereof, and / or the like.

[0125] In some non-limiting embodiments or aspects, the pretrained machine learning model may have been trained based on training data, which may include a plurality of data records. Each data record may include a plurality of features. For example, each data record may be represented as a vector (or tensor) of values, and each respective value may be associated with a respective feature.5Y77189.DOCX Page 34 of 69Attorney Docket No. 08223-2405237 (8270WG01 )

[0126] In some non-limiting embodiments or aspects, receiving the pretrained machine learning model may include development environment system 102 retrieving the pretrained machine learning model from storage (e.g., data storage system 104).

[0127] In some non-limiting embodiments or aspects, receiving the pretrained machine learning model may include development environment system 102 receiving the pretrained machine learning model from production environment system 106 (or retrieving the pretrained machine learning model from data storage system 104 based on receiving a communication from production environment system 106). For example, the pretrained machine learning model may be deployed by production environment system 106 in a production environment to generate outputs based on incoming data records, as described herein. Production environment system 106 may detect at least one anomaly (e.g., a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a first distribution of the incoming data records for a first time period with respect to a second distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, any combination thereof, and / or the like) in the production environment, as described herein. Production environment system 106 may communicate with development environment system 102 based on detecting the anomaly.

[0128] For the purpose of illustration, with reference to FIG. 5A and with continued reference to FIG. 2A, FIG. 5A shows an example pretrained machine learning model 510 of an example implementation 500 of systems and methods for improved model training using reinforcement learning and feature dropout, according to some nonlimiting embodiments or aspects.

[0129] As shown in FIG. 5A, input data 501 may include a plurality of features (e.g., Feature 1 , Feature 2, Feature 3, etc.). Input layer 51 1 of pretrained machine learning model 510 may be configured to receive input data 501. For example, if each data record of input data 501 is a vector of values that each correspond to a respective feature, as described herein, input layer 511 may include a node for each feature (e.g., a node for each value of the vector).

[0130] In some non-limiting embodiments or aspects, hidden layer block 513 may include at least one hidden layer of pretrained machine learning model 510. For example, hidden layer block 513 may include a plurality of neural network layers,5Y77189.DOCX Page 35 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) transformer layers, attention layers, convolutional layers, recurrent layers, and / or the like. In some non-limiting embodiments or aspects, each data record of input data 501 may be received at input layer 51 1 and forward propagated through hidden layer block 513 to produce an output 515 (e.g., prediction y).

[0131] Referring again to FIG. 2A, at step 214, method 200a may include initializing a second (e.g., tunable) machine learning model (e.g., including a feature selection layer). For example, development environment system 102 may initialize a second machine learning model, as described herein.

[0132] In some non-limiting embodiments or aspects, the second machine learning model may include a second input layer, a feature selection layer, and a second plurality of hidden layers. The second input layer (of the second machine learning model) may be initialized based on the first input layer (of the first / pretrained machine learning model). For example, parameters of the first input layer may be copied as the second input layer (e.g., the second input layer initially may be a copy of the first input layer). The second plurality of hidden layers (of the second machine learning model) may be initialized based on the first plurality of hidden layers (of the first / pretrained machine learning model). For example, parameters of the first plurality of hidden layers may be copied as the second plurality of hidden layers (e.g., the second plurality of hidden layers initially may be a copy of the first plurality of hidden layers).

[0133] In some non-limiting embodiments or aspects, the feature selection layer may include a plurality of weights, including a respective weight for each respective feature of the plurality of features (e.g., a weight for each node of the input layer). Each respective weight may be associated with a respective dropout rate for the respective feature.

[0134] For the purpose of illustration, with reference to FIG. 5B and with continued reference to FIG. 2A, FIG. 5B shows a schematic diagram of example implementation 500 of systems and methods for improved model training using reinforcement learning and feature dropout, according to some non-limiting embodiments or aspects.

[0135] As shown in FIG. 5B, second machine learning model 520 may include input layer 521 , feature selection layer 522, and hidden layer block 523. In some non-limiting embodiments or aspects, input layer 521 and hidden layer block 523 of second machine learning model 520 may be copies of input layer 51 1 and hidden layer block 513, respectively, of pretrained machine learning model 510.5Y77189.DOCX Page 36 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0136] Feature selection layer 522 may have parameters 0, which may be a vector (or tensor) of weights 0t. Each weight 0tmay be associated with the dropout rate for the ith feature (e.g., Feature i). For example, the dropout rate may be 1 - 0t(e.g., the ith feature has a 1 - 0tprobability to be dropped). For the purpose of illustration, for each element xtof a data record x received in input layer 521 , the intermediate output x't of feature selection layer 522 may be represented as x't= Bernoulli O^xt, where Bernoulli 0i) is the Bernoulli distribution function that has a value of 1 with probability 0tand a value of 0 with probability (1 -0t).

[0137] Referring again to FIG. 2A, at step 216, method 200a may include adjusting the weights of the feature selection layer based on at least one reinforcement learning reward. For example, development environment system 102 may adjust the weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward, as described herein.

[0138] In some non-limiting embodiments or aspects, development environment system 102 may adjust the plurality of weights of the feature selection layer of the second machine learning model based on the training data and the at least one reinforcement learning reward to provide an adjusted plurality of weights.

[0139] In some non-limiting embodiments or aspects, the at least one reinforcement learning reward may include at least two of: a feature importance reward, a loss-based reward, a divergence-based reward, and / or the like. For example, the reinforcement learning reward(s) may include all of the feature importance reward, the loss-based reward, and divergence-based reward.

[0140] In some non-limiting embodiments or aspects, the feature importance reward may be based on a Pearson rank correlation and a reference feature importance ranking / (x). For example, the reference feature importance ranking / (x) may be determined based on any suitable technique for feature importance ranking. For the purpose of illustration, the reference feature importance ranking / (x) may be determined based on at least one of Shapley values, Shapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), submodular pick (SP) LIME, tree-based models, random forest, and / or the like. In some non-limiting embodiments or aspects, the feature importance reward may reward a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights 0tcorrelates with the reference feature importance ranking / (x).5Y77189.DOCX Page 37 of 69Attorney Docket No. 08223-2405237 (8270W001 )

[0141] For the purpose of illustration, the feature importance reward may be represented by the following equation:where rgis the feature importance reward, p is the Pearson rank correlation function, and rank is the rank function. For example, this equation is positive, so a greater correlation may be associated with a greater reward for the reinforcement learning agent.

[0142] In some non-limiting embodiments or aspects, the loss-based reward may include any suitable loss function for training a machine learning mode. For example, the loss-based reward may be based on a cross-entropy (CE) loss. In some nonlimiting embodiments or aspects, the loss-based reward may be based on the loss function used to pretrain the pretrained model. In some non-limiting embodiments or aspects, the loss-based reward (e.g., CE loss) may be determined based on the output y' generated by second machine learning model 520 based on the input data record x. In some non-limiting embodiments or aspects, the loss-based reward may penalize the reinforcement learning agent for a loss.

[0143] For the purpose of illustration, the loss-based reward may be represented by the following equation:where rgis the loss-based reward, ygtis the ground truth labels, andtunedis the output distribution of the second machine learning model. For example, the negative sign in the equation may result in a greater penalty for the reinforcement learning agent as the loss increases.

[0144] In some non-limiting embodiments or aspects, the divergence-based reward may be determined based on any suitable divergence metric to measure divergence between an output y generated by pretrained machine learning model 510 based on an input data record x and an output y' generated by second machine learning model 520 based on the input data record x. For example, the divergence-based reward may be based on a Kullback-Leibler (KL) divergence (e.g., between y and y'). In some non-limiting embodiments or aspects, the divergence-based reward may penalize the reinforcement learning agent for a divergence.

[0145] For the purpose of illustration, the divergence-based reward may be represented by the following equation:5Y77189.DOCX Page 38 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) r0(y'|%) = ^-KLDKL (^tuned (.y' \x, 0) \ \Tlbase (yW) where reis the divergence-based reward,KLis a coefficient, DKLis the KL divergence function,tunedis the output distribution of the second machine learning model, nbaseis the output distribution of the pretrained machine learning model. For example, the negative sign in the equation may result in a greater penalty for the reinforcement learning agent as the divergence increases.

[0146] In some non-limiting embodiments or aspects, a total reward may be determined based on the feature importance reward, the loss-based reward, and divergence-based reward. For example, the total reward may be the sum of the feature importance reward, the loss-based reward, and divergence-based reward.

[0147] In some non-limiting embodiments or aspects, development environment system 102 (e.g., a reinforcement learning agent thereof) may adjust the plurality of weights 0tto at least one of increase the total reward (e.g., exploitation), randomly adjust at least one weight 0tof the plurality of weights (e.g., exploration, i.e., take random action to explore the solution space), or any combination thereof. For example, development environment system 102 (e.g., a reinforcement learning agent thereof) may use an Epsilon-Greedy algorithm to balance exploration and exploitation (e.g., action at time t: increase the total reward with probability E, take other (e.g., random) action with probability 1 - E).

[0148] In some non-limiting embodiments or aspects, only the parameters 0 (e.g., the plurality of weights 0tof feature selection layer 522) are updated by the reinforcement learning agent. For example, the parameters of hidden layer block 523 may remain unchanged between iterations of the reinforcement agent. Additionally, pretrained model 510 may not change between iterations of the reinforcement agent.

[0149] For the purpose of illustration, referring again to FIG. 5B and with continued reference to FIG. 2A, the output 525 y' and the parameters 0 (e.g., the plurality of weights 0tof feature selection layer 522) are used as inputs to feature importance reward 530, loss-based reward 540 (e.g., CE loss), and divergence-based reward 550 (e.g., KL loss), as described herein. Additionally, the output 515 y of pretrained model 510 is used as input for divergence-based reward 550 (e.g., KL loss), as described herein.

[0150] In some non-limiting embodiments or aspects, feature importance reward 530 may use feature layer 531 (e.g., which may be the same as input layer 521 and / or5Y77189.DOCX Page 39 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) input layer 51 1 ) as input to feature importance block 533, which may be used to determine (e.g., generate) the reference feature importance ranking / (%), as described herein.

[0151] In some non-limiting embodiments or aspects, total reward 560 may be determined based on a sum of feature importance reward 530, loss-based reward 540 (e.g., CE loss), and divergence-based reward 550 (e.g., KL loss), as described herein.

[0152] In some non-limiting embodiments or aspects, reinforcement learning agent 570 (e.g., RL Agent) may take an action based on the total reward, as described herein. For example, reinforcement learning agent 570 may update the parameters 0 (e.g., the plurality of weights 0tof feature selection layer 522) to increase reward (e.g., with probability E in an Epsilon-Greedy algorithm) or explore (e.g., randomly adjust) the parameters 0 (e.g., with probability 1 - E in an Epsilon-Greedy algorithm).

[0153] In some non-limiting embodiments or aspects, reinforcement learning agent 570 may repeat the process until a termination condition is satisfied. For example, a termination condition may include that the marginal change in reward between successive iterations satisfies a threshold (e.g., is less than a threshold) or that a target (e.g. maximum) number of iterations has been reached. In some non-limiting embodiments or aspects, each iteration of the reinforcement learning agent may include forward propagating each data record of input data 501 through input layer 521 , feature selection layer 522 (with its current parameters / weights), and hidden layer block 523 so that outputs 525 may be used to determine the feature importance reward 530, loss-based reward 540 (e.g., CE loss), divergence-based reward 550 (e.g., KL loss), and total reward 560 for each iteration.

[0154] In some non-limiting embodiments or aspects, only the parameters 0 (e.g., the plurality of weights 0tof feature selection layer 522) are updated by reinforcement learning agent 570. For example, the parameters of hidden layer block 523 may be frozen (e.g., remain unchanged) between iterations of reinforcement agent 570. Additionally, pretrained model 510 may be frozen (e.g., remain unchanged) between iterations of reinforcement agent 570.

[0155] Referring again to FIG. 2A, at step 218, method 200a may include retraining other layers (e.g., hidden layer(s)) of the second machine learning model. For example, development environment system 102 may retrain the hidden layer(s) of the second machine learning model, as described herein.5Y77189.DOCX Page 40 of 69Attorney Docket No. 08223-2405237 (8270WG01 )

[0156] In some non-limiting embodiments or aspects, the weights of the feature selection layer may remain unchanged during retraining of the hidden layer(s). For example, retraining may include receiving each data record of training data at the input layer of the second machine learning model to be forward propagated through the feature selection layer (which will dropout certain features based on the weights thereof) and hidden layers of the second machine learning model to produce an output. A loss may be determined based on output. In some non-limiting embodiments or aspects, the same loss function used for the loss-based reward may be used to determine the loss for retraining. Additionally, or alternatively, any suitable loss function may be used for retraining, such as an error, a mean error, a mean squared error (MSE), a CE loss, any combination thereof, and / or the like. The parameters of the hidden layer(s) may be updated based on the loss. For example, development environment system 102 may update (e.g., adjust) the parameters of the hidden layers of the second machine learning model based on back propagation (e.g., of the loss(es)), gradient calculations (e.g., based on the loss(es)), any combination thereof, and / or the like.

[0157] For the purpose of illustration, with reference to FIG. 50 and with continued reference to FIG. 2A, FIG. 50 shows an example tuned machine learning model 520 of an example implementation 500 of systems and methods for improved model training using reinforcement learning and feature dropout, according to some nonlimiting embodiments or aspects.

[0158] In some non-limiting embodiments or aspects, the weights 0tof feature selection layer 522 may be frozen (e.g., remain unchanged) during retraining of hidden layer block 523. For example, retraining may include receiving each data record of input data 501 at input layer 521 to be forward propagated through feature selection layer 522 (which will dropout certain features based on the weights thereof) and hidden layer block 523 to produce an output 525. A loss may be determined based on output. In some non-limiting embodiments or aspects, the same loss function used for the loss-based reward may be used to determine the loss for retraining (e.g., a CE loss). The parameters of hidden layer block 523 may be updated based on the loss (e.g., based on back propagation, gradient calculations, any combination thereof, and / or the like), as described herein.

[0159] In some non-limiting embodiments or aspects, second machine learning model 520 having input layer 521 , feature selection layer 522 having the adjusted5Y77189.DOCX Page 41 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) plurality of weights, and hidden layer block 523 having the adjusted parameters may be referred to as a tuned machine learning model. In some non-limiting embodiments or aspects, the tuned machine learning model may be deployed in the production environment (e.g., in place of the pretrained model). For example, production environment system 106 may deploy the second machine learning model in the production environment, as described herein. In some non-limiting embodiments or aspects, development environment system 102 may communicate the tuned machine learning model to production environment system 106. Additionally, or alternatively, production environment system 106 may retrieve the tuned machine learning model (e.g., from data storage system 104). For example, production environment system 106 may retrieve the tuned machine learning model based on receiving a communication from development environment system 102.

[0160] Referring now to FIG. 2B, shown is a flow diagram for an example method for 200b, according to some non-limiting embodiments or aspects. The steps shown in FIG. 2B are for example purposes only. It will be appreciated that additional, fewer, different, and / or a different order of steps may be used in some non-limiting embodiments or aspects. In some non-limiting embodiments or aspects, a step may be automatically performed in response to performance and / or completion of a prior step. In some non-limiting embodiments or aspects, one or more of the steps of method 200b may be performed (e.g., completely, partially, and / or the like) by development environment system 102 (e.g., one or more devices of development environment system 102). In some non-limiting embodiments or aspects, one or more of the steps of method 200b may be performed (e.g., completely, partially, and / or the like) by another system, another device, another group of systems, or another group of devices, separate from or including development environment system 102, such as data storage system 104, production environment system 106, and / or the like.

[0161] As shown in FIG. 2B, at step 202, method 200b may include deploying a first (e.g., pretrained) machine learning model in a production environment. For example, production environment system 106 may deploy a pretrained machine learning model in a production environment, as described herein.

[0162] For example, production environment system 106 may deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records. The pretrained machine learning model may include a first input layer and a first plurality of hidden layers. The pretrained machine learning model5Y77189.DOCX Page 42 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) may have been trained based on training data including a plurality of data records. Each data record of the plurality of data records may include a plurality of features.

[0163] As shown in FIG. 2B, at step 204, method 200b may include detecting at least one anomaly in the production environment. For example, production environment system 106 may detect at least one anomaly in the production environment, as described herein.

[0164] For example, the at least one anomaly may include at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a first distribution of the incoming data records for a first time period with respect to a second distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, any combination thereof, and / or the like.

[0165] As shown in FIG. 2B, at step 212, method 200b may include receiving a first (e.g., pretrained) machine learning model. For example, development environment system 102 may receive a pretrained machine learning model, as described herein. In some non-limiting embodiments or aspects, step 212 as shown in FIG. 2B may be the same as or similar to step 212 shown in FIG. 2A.

[0166] For example, based on production environment system 106 detecting the at least one anomaly in the production environment, development environment system 102 may receive the pretrained machine learning model in a development environment.

[0167] As shown in FIG. 2B, at step 214, method 200b may include initializing a second (e.g., tunable) machine learning model (e.g., including a feature selection layer). For example, development environment system 102 may initialize a second machine learning model, as described herein. In some non-limiting embodiments or aspects, step 214 as shown in FIG. 2B may be the same as or similar to step 214 shown in FIG. 2A.

[0168] For example, development environment system 102 may initialize a second machine learning model including a second input layer, a feature selection layer, and a second plurality of hidden layers. The second input layer may be initialized based on the first input layer of the pretrained machine learning mode. The second plurality of hidden layers may be initialized based on the first plurality of hidden layers of the pretrained machine learning model. The feature selection layer may include a plurality5Y77189.DOCX Page 43 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) of weights including a respective weight for each respective feature of the plurality of features. Each respective weight of the plurality of weights may be associated with a respective dropout rate for the respective feature of the plurality of features.

[0169] As shown in FIG. 2B, at step 216, method 200b may include adjusting the weights of the feature selection layer based on at least one reinforcement learning reward. For example, development environment system 102 may adjust the weights of the features selection layer of the second machine learning model based on at least one reinforcement learning reward, as described herein. In some non-limiting embodiments or aspects, step 216 as shown in FIG. 2B may be the same as or similar to step 216 shown in FIG. 2A.

[0170] For example, development environment system 102 may adjust the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, any combination thereof, and / or the like to provide an adjusted plurality of weights.

[0171] As shown in FIG. 2B, at step 218, method 200b may include retraining other layers (e.g., hidden layer(s)) of the second machine learning model. For example, development environment system 102 may retrain the hidden layer(s) of the second machine learning model, as described herein. In some non-limiting embodiments or aspects, step 218 as shown in FIG. 2B may be the same as or similar to step 218 shown in FIG. 2A.

[0172] For example, development environment system 102 may retrain the second plurality of hidden layers based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, any combination thereof, and / or the like to provide adjusted parameters of the second plurality of hidden layers.

[0173] In some non-limiting embodiments or aspects, a tuned machine learning model may include the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.

[0174] In some non-limiting embodiments or aspects, development environment system 102 may communicate the tuned machine learning model to production environment system 106.5Y77189.DOCX Page 44 of 69Attorney Docket No. 08223-2405237 (8270WG01 )

[0175] As shown in FIG. 2B, at step 222, method 200b may include deploying the second (e.g., tuned) machine learning model in the production environment. For example, production environment system 106 may deploy the tuned machine learning model in the production environment, as described herein.

[0176] For example, production environment system 106 may deploy the tuned machine learning model in the production environment to generate further outputs based on further incoming data records.

[0177] Referring now to FIG. 3, depicted is a diagram of an example payment processing network 300, according to non-limiting embodiments or aspects. In some non-limiting embodiments or aspects, payment processing network 300 may be used in conjunction with the systems, methods, and / or computer program products described herein, and / or the systems, methods, and / or computer program products described herein may be implemented in payment processing network 300. As shown in FIG. 3, payment processing network 300 may include transaction processing system 301 , payment gateway system 302, merchant system 304, issuer system 306, acquirer system 308, and / or consumer device 310. In some non-limiting embodiments or aspects, each of development environment system 102, data storage system 104, and / or production environment system 106 of FIG. 1 may be implemented by (e.g., part of) transaction processing system 301. For example, development environment system 102 and production environment system 106 may be logically or physically separate subsystems of transaction processing system 301. In some non-limiting embodiments or aspects, at least one of development environment system 102, data storage system 104, and / or production environment system 106 of FIG. 1 may be implemented by (e.g., part of) another system, another device, another group of systems, or another group of devices, separate from or including transaction processing system 301 , such as merchant system 304, issuer system 306, acquirer system 308, consumer device 310, and / or the like. For example, development environment system 102, data storage system 104, and / or production environment system 106 may be implemented by (e.g., part of) issuer system 306. Additionally, or alternatively, for example, data storage system 104 and production environment system 106 may be implemented by (e.g., part of) at least one of payment gateway system 302, merchant system 304, issuer system 306, acquirer system 308, and / or consumer device 310, while development environment system 102 may be implemented by (e.g., part of) transaction processing system 301 .5Y77189.DOCX Page 45 of 69Attorney Docket No. 08223-2405237 (8270WG01 )

[0178] Transaction processing system 301 may include one or more devices capable of receiving information from and / or communicating information to payment gateway system 302, merchant system 304, issuer system 306, acquirer system 308, consumer device 310, and / or the like (e.g., directly, indirectly, via a public and / or private communication network connection, and / or the like). For example, as shown in FIG. 3, transaction processing system 301 may be in communication with one or more issuer systems (e.g., issuer system 306), one or more acquirer systems (e.g., acquirer system 308), and / or one or more payment gateway systems (e.g., payment gateway system 302). Although only a single issuer system 306, single acquirer system 308, and single payment gateway system 302 are shown, it will be appreciated that transaction processing system 301 may be in communication with a plurality of issuer systems, a plurality of acquirer systems, and / or a plurality of payment gateways. In some non-limiting embodiments or aspects, transaction processing system 301 may include a computing device, such as a server (e.g., a transaction processing server), a group of servers, and / or other like devices. In some non-limiting embodiments or aspects, transaction processing system 301 may be in communication with a data storage device, which may be local or remote to transaction processing system 301. In some non-limiting embodiments or aspects, transaction processing system 301 may be capable of receiving information from, storing information in, communicating information to, or searching information stored in the data storage device. In some non-limiting embodiments or aspects, transaction processing system 301 may be associated with a transaction service provider, as described herein. In some nonlimiting embodiments or aspects, transaction processing system 301 may also operate as an issuer system 306 such that both transaction processing system 301 and issuer system 306 are a single system and / or controlled by a single entity.

[0179] Payment gateway system 302 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 301 , merchant system 304, issuer system 306, acquirer system 308, consumer device 310, and / or the like (e.g., directly, indirectly, via a public and / or private communication network connection, and / or the like). For example, as shown in FIG. 3, payment gateway system 302 may be in communication with one or more merchant systems (e.g., merchant system 304), one or more acquirer systems (e.g., acquirer system 308), and / or one or more transaction processing systems (e.g., transaction processing system 301 ). Although only a single merchant system 304,5Y77189.DOCX Page 46 of 69Attorney Docket No. 08223-2405237 (8270W001 ) single acquirer system 308, and single transaction processing system 301 are shown, it will be appreciated that payment gateway system 302 may be in communication with a plurality of merchant systems, a plurality of acquirer systems, and / or a plurality of transaction processing systems. In some non-limiting embodiments or aspects, payment gateway system 302 may include a computing device, such as a server, a group of servers, and / or other like devices. In some non-limiting embodiments or aspects, payment gateway system 302 may be associated with a payment gateway, as described herein.

[0180] Merchant system 304 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 301 , payment gateway system 302, issuer system 306, acquirer system 308, consumer device 310, and / or the like (e.g., directly, indirectly, via a public and / or private communication network connection, and / or the like). For example, as shown in FIG. 3, merchant system 304 may be in communication with one or more payment gateway systems (e.g., payment gateway system 302), one or more acquirer systems (e.g., acquirer system 308), and / or one or more consumer devices (e.g., consumer device 310). Although only a single payment gateway system 302, single acquirer system 308, and single consumer device 310 are shown, it will be appreciated that merchant system 304 may be in communication with a plurality of payment gateway systems, a plurality of acquirer systems, and / or a plurality of consumer devices. In some non-limiting embodiments or aspects, merchant system 304 may include a computing device, such as a server, a group of servers, a client device, a group of client devices, a POS device, a POS system, computers, computer systems, peripheral devices, and / or other like devices. In some non-limiting embodiments or aspects, merchant system 304 may be associated with a merchant, as described herein. In some non-limiting embodiments or aspects, merchant system 304 may include a device capable of receiving information from and / or communicating information to consumer device 310 via a short-range communication connection (e.g., an NFC communication connection, an RFID communication connection, a Bluetooth® communication connection, a Zigbee® communication connection, and / or the like) with consumer device 310 and / or the like. In some non-limiting embodiments or aspects, merchant system 304 may include one or more client devices. For example, merchant system 304 may include a client device that allows a merchant to communicate information to transaction processing system 301 (e.g., via at least one5Y77189.DOCX Page 47 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) of acquirer system 308 and / or payment gateway system 302). In some non-limiting embodiments or aspects, merchant system 304 (e.g., a client device thereof, a POS device thereof, and / or the like) may also operate as a payment gateway system 302 such that both merchant system 304 and payment gateway system 302 are a single system and / or controlled by a single entity.

[0181] Issuer system 306 may include one or more devices capable of receiving information and / or communicating information to transaction processing system 301 , payment gateway system 302, merchant system 304, acquirer system 308, consumer device 310, and / or the like (e.g., directly, indirectly, via a public and / or private communication network connection, and / or the like). For example, as shown in FIG. 3, issuer system 306 may be in communication with one or more transaction processing systems (e.g., transaction processing system 301 ) and / or one or more consumer devices (e.g., consumer device 310). Although only a single transaction processing system 301 and a single consumer device 310 are shown, it will be appreciated that issuer system 306 may be in communication with a plurality of transaction processing systems and / or a plurality of consumer devices 310. In some non-limiting embodiments or aspects, issuer system 306 may include a computing device, such as a server, a group of servers, and / or other like devices. In some nonlimiting embodiments or aspects, issuer system 306 may be associated with an issuer institution, as described herein. For example, issuer system 306 may be associated with an issuer institution that issued a credit account, a debit account, a credit card, a debit card, a payment device, and / or the like to a user associated with consumer device 310.

[0182] Acquirer system 308 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 301 , payment gateway system 302, merchant system 304, issuer system 306, consumer device 310, and / or the like (e.g., directly, indirectly, via a public and / or private communication network connection, and / or the like). For example, as shown in FIG. 3, acquirer system 308 may be in communication with one or more transaction processing systems (e.g., transaction processing system 301 ), one or more payment gateway systems (e.g., payment gateway system 302), and / or one or more merchant systems (e.g., merchant system 304). Although only a single transaction processing system 301 , a single payment gateway system 302, and a single merchant system 304 are shown, it will be appreciated that acquirer system 308 may be in5Y77189.DOCX Page 48 of 69Attorney Docket No. 08223-2405237 (8270W001 ) communication with a plurality of transaction processing systems, a plurality of payment gateway systems, and / or a plurality of merchant systems. In some nonlimiting embodiments or aspects, acquirer system 308 may include a computing device, such as a server, a group of servers, and / or other like devices. In some nonlimiting embodiments or aspects, acquirer system 308 may be associated with an acquirer institution, as described herein.

[0183] Consumer device 310 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 301 , payment gateway system 302, merchant system 304, issuer system 306, acquirer system 308, and / or the like (e.g., directly, indirectly, via a public and / or private communication network connection, and / or the like). For example, as shown in FIG. 3, consumer device 310 may be in communication with one or more merchant systems (e.g., merchant system 304) and / or one or more issuer systems (e.g., issuer system 306). Although only a single merchant system 304 and a single issuer system 306 are shown, it will be appreciated that consumer device 310 may be in communication with a plurality of merchant systems and / or a plurality of issuer systems. In some nonlimiting embodiments or aspects, consumer device 310 may be associated with a user to whom a credit account, a debit account, a credit card, a debit card, a payment device, and / or the like has been issued. In some non-limiting embodiments or aspects, consumer device 310 may include a computing device, such as a computer, a portable computer, a laptop computer, a tablet computer, a mobile device, a cellular phone, a smartphone, a wearable device (e.g., watches, glasses, lenses, clothing, and / or the like), a PDA, a client device, and / or other like devices. In some non-limiting embodiments or aspects, consumer device 310 may include a payment device, as described herein. In some non-limiting embodiments or aspects, consumer device 310 may include a device capable of receiving information from and / or communicating information to other consumer devices 310 (e.g., directly, indirectly, via a public and / or private communication network connection, a short-range communication connection, and / or the like). In some non-limiting embodiments or aspects, consumer device 310 may include a device capable of receiving information from and / or communicating information to merchant system 304 via a short-range communication connection (e.g., an NFC communication connection, an RFID communication connection, a Bluetooth® communication connection, a Zigbee® communication connection, and / or5Y77189.DOCX Page 49 of 69Attorney Docket No. 08223-2405237 (8270W001 ) the like) with merchant system 304 and / or the like. In some non-limiting embodiments or aspects, consumer device 310 may include a client device.

[0184] In some non-limiting embodiments or aspects, transaction processing system 301 may communicate with merchant system 304 directly (e.g., via a public and / or private communication network connection and / or the like). Additionally, or alternatively, transaction processing system 301 may communicate with merchant system 304 through payment gateway 302 and / or acquirer system 308. In some nonlimiting embodiments or aspects, acquirer system 308 associated with merchant system 304 may operate as payment gateway 302 to facilitate the communication of transaction messages (e.g., authorization requests) from merchant system 304 to transaction processing system 301. In some non-limiting embodiments or aspects, merchant system 304 may communicate with payment gateway 302 directly (e.g., via a public and / or private communication network connection and / or the like). For example, merchant system 304 that includes a physical POS device may communicate with payment gateway 302 through a public or private network to conduct card-present transactions. As another example, merchant system 304 that includes a server (e.g., a web server) may communicate with payment gateway 302 through a public or private network, such as the Internet, to conduct card-not-present transactions.

[0185] For the purpose of illustration, processing a transaction (e.g., a payment transaction) may include generating a transaction message (e.g., authorization request and / or the like) based on an account identifier of a customer (e.g., accountholder associated with consumer device 310 and / or the like) and / or transaction data associated with the transaction. For example, merchant system 304 (e.g., a client device of merchant system 304, a POS device of merchant system 304, and / or the like) may initiate the transaction, e.g., by generating an authorization request (e.g., in response to receiving the account identifier from a payment device and / or a portable financial device of the customer and / or the like). Merchant system 304 may communicate the authorization request to payment gateway 302 and / or acquirer system 308. In some non-limiting embodiments or aspects, payment gateway 302 may communicate the authorization request to acquirer system 308 and / or transaction processing system 301 . Additionally, or alternatively, acquirer system 308 (and / or payment gateway 302) may communicate the authorization request to transaction processing system 301. After receiving the authorization request from5Y77189.DOCX Page 50 of 69Attorney Docket No. 08223-2405237 (8270W001 ) merchant system 304 that identifies the account identifier of the customer (e.g., the accountholder associated with consumer device 310 and / or the account identifier), transaction processing system 301 may communicate the authorization request to issuer system 306 (e.g., the issuer system that issued the payment device and / or account identifier). Issuer system 306 may determine an authorization decision (e.g., approve, deny, and / or the like) based on the authorization request, and / or issuer system 306 may generate an authorization response based on the authorization decision and / or the authorization request. Issuer system 306 may communicate the authorization response to transaction processing system 301 . Transaction processing system 301 may communicate the authorization response to acquirer system 308 and / or payment gateway 302. In some non-limiting embodiments or aspects, acquirer system 308 may communicate the authorization response to payment gateway 302 and / or merchant system 304. Additionally, or alternatively, payment gateway 302 (and / or acquirer system 308) may communicate the authorization response to merchant system 304.

[0186] In some non-limiting embodiments or aspects, transaction processing system 301 and / or issuer system 306 may include at least one machine learning model (e.g., at least one of a fraud detection model, a risk detection model, a transaction authorization model, a clearing and / or settlement prediction model, a credit approval model, a product recommendation model, a classifier model, an anomaly detection model, an authentication model, any combination thereof, and / or the like). For example, the machine learning model(s) may include at least one of a neural network, a CNN, an RNN, an autoencoder (or portion thereof), an LSTM, a bidirectional LSTM, an attention network, a transformer, a tree-based model, a decision tree model, a random forest model, a gradient boosting model, an XGBoost model, any combination thereof, and / or the like, as described herein. Transaction processing system 301 and / or issuer system 306 may perform at least one task (e.g., generate a prediction and / or classification) based on the authorization request and the machine learning model(s). For example, performing the task(s) may include generating at least one prediction associated with fraud detection, risk detection, transaction authorization, clearing and / or settlement, credit approval, product recommendation, classification, anomaly detection, authentication, any combination thereof, and / or the like. In some non-limiting embodiments or aspects, transaction processing system 301 may communicate at least one message based on performing5Y77189.DOCX Page 51 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) the task (e.g., generating the prediction and / or classification) to issuer system 306 (e.g., along with the authorization request). In some non-limiting embodiments or aspects, issuer system 306 may determine the authorization decision (e.g., approve, deny, and / or the like) based on the authorization request and the performance of the task (e.g., generation of the prediction and / or classification). In some non-limiting embodiments or aspects, the machine learning model(s) may be trained (e.g., tuned) based on reinforcement learning and feature dropout, as described herein.

[0187] For the purpose of illustration, clearing and / or settlement of a transaction may include generating a message (e.g., clearing message and / or the like) based on an account identifier of a customer (e.g., associated with consumer device 310 and / or the like) and / or transaction data associated with the transaction. For example, merchant system 304 may generate at least one clearing message (e.g., a plurality of clearing messages, a batch of clearing messages, and / or the like). Merchant system 304 may communicate the clearing message(s) to acquirer system 308 (and / or payment gateway 302, which may communicate the clearing message(s) to acquirer system 308). Acquirer system 308 may communicate the clearing message(s) to transaction processing system 301. Transaction processing system 301 may communicate the clearing message(s) to issuer system 306. Issuer system 306 may generate at least one settlement message based on the clearing message(s). In some non-limiting embodiments or aspects, issuer system 306 may communicate the settlement message(s) and / or funds to transaction processing system 301 (and / or a settlement bank system associated with transaction processing system 301 ), and transaction processing system 301 (and / or the settlement bank system) may communicate the settlement message(s) and / or funds to acquirer system 308. Additionally, or alternatively, issuer system 306 may communicate the settlement message(s) and / or funds to acquirer system 308. In some non-limiting embodiments or aspects, acquirer system 308 may communicate settlement message(s) and / or funds to merchant system 304 (and / or an account associated with merchant system 304).

[0188] The systems and / or devices of FIG. 3 may communicate via one or more wired and / or wireless communication networks. For example, the communication network(s) may include a cellular network (e.g., a long-term evolution (LTE®) network, a third generation (3G) network, a fourth generation (4G) network, a fifth generation (5G) network, a code division multiple access (CDMA) network, and / or the like), a5Y77189.DOCX Page 52 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) public land mobile network (PLMN), a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), a telephone network (e.g., the public switched telephone network (PSTN)), a private network (e.g., a private network associated with a transaction service provider), an ad hoc network, an intranet, the Internet, a fiber optic-based network, a cloud computing network, and / or the like, and / or a combination of these or other types of networks.

[0189] The number and arrangement of systems, devices, and / or networks shown in FIG. 3 are provided as an example. There may be additional systems, devices, and / or networks; fewer systems, devices, and / or networks; different systems, devices, and / or networks; and / or differently arranged systems, devices, and / or networks than those shown in FIG. 3. Furthermore, two or more systems or devices shown in FIG. 3 may be implemented within a single system or device, or a single system or device shown in FIG. 3 may be implemented as multiple, distributed systems or devices. Additionally, or alternatively, a set of systems (e.g., one or more systems) or a set of devices (e.g., one or more devices) of payment processing network 300 may perform one or more functions described as being performed by another set of systems or another set of devices of payment processing network 300.

[0190] Referring now to FIG. 4, shown is a diagram of example components of device 400 according to non-limiting embodiments. Device 400 may correspond to at least one of development environment system 102, data storage system 104, production environment system 106, and / or communication network 1 12 of FIG. 1 and / or at least one of transaction processing system 301 , payment gateway system 302, merchant system 304, issuer system 306, acquirer system 308, and / or consumer device 310 in FIG. 3, as an example. In some non-limiting embodiments or aspects, such systems or devices may include at least one device 400 and / or at least one component of device 400. The number and arrangement of components shown are provided as an example. In some non-limiting embodiments or aspects, device 400 may include additional components, fewer components, different components, or differently arranged components than those shown. Additionally, or alternatively, a set of components (e.g., one or more components) of device 400 may perform one or more functions described as being performed by another set of components of device 400.

[0191] As shown in FIG. 4, device 400 may include bus 402, processor 404, memory 406, storage component 408, input component 410, output component 412,5Y77189.DOCX Page 53 of 69Attorney Docket No. 08223-2405237 (8270WG01 ) and communication interface 414. Bus 402 may include a component that permits communication among the components of device 400. In some non-limiting embodiments or aspects, processor 404 may be implemented in hardware, firmware, or a combination of hardware and software. For example, processor 404 may include a processor (e.g., a central processing unit (CPU), a graphics processing unit (GPU), an accelerated processing unit (APU), etc.), a microprocessor, a digital signal processor (DSP), and / or any processing component (e.g., a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), etc.) that can be programmed to perform a function. Memory 406 may include random access memory (RAM), read only memory (ROM), and / or another type of dynamic or static storage device (e.g., flash memory, magnetic memory, optical memory, etc.) that stores information and / or instructions for use by processor 404.

[0192] With continued reference to FIG. 4, storage component 408 may store information and / or software related to the operation and use of device 400. For example, storage component 408 may include a hard disk (e.g., a magnetic disk, an optical disk, a magneto-optic disk, a solid-state disk, etc.) and / or another type of computer-readable medium. Input component 410 may include a component that permits device 400 to receive information, such as via user input (e.g., a touch screen display, a keyboard, a keypad, a mouse, a button, a switch, a microphone, etc.). Additionally, or alternatively, input component 410 may include a sensor for sensing information (e.g., a global positioning system (GPS) component, an accelerometer, a gyroscope, an actuator, etc.). Output component 412 may include a component that provides output information from device 400 (e.g., a display, a speaker, one or more light-emitting diodes (LEDs), etc.). Communication interface 414 may include a transceiver-like component (e.g., a transceiver, a separate receiver and transmitter, etc.) that enables device 400 to communicate with other devices, such as via a wired connection, a wireless connection, or a combination of wired and wireless connections. Communication interface 414 may permit device 400 to receive information from another device and / or provide information to another device. For example, communication interface 414 may include an Ethernet interface, an optical interface, a coaxial interface, an infrared interface, a radio frequency (RF) interface, a universal serial bus (USB) interface, a Wi-Fi® interface, a cellular network interface, and / or the like.5Y77189.DOCX Page 54 of 69Attorney Docket No. 08223-2405237 (8270WG01 )

[0193] Device 400 may perform one or more processes described herein. Device 400 may perform these processes based on processor 404 executing software instructions stored by a computer-readable medium, such as memory 406 and / or storage component 408. A computer-readable medium may include any non-transitory memory device. A memory device includes memory space located inside of a single physical storage device or memory space spread across multiple physical storage devices. Software instructions may be read into memory 406 and / or storage component 408 from another computer-readable medium or from another device via communication interface 414. When executed, software instructions stored in memory 406 and / or storage component 408 may cause processor 404 to perform one or more processes described herein. Additionally, or alternatively, hardwired circuitry may be used in place of or in combination with software instructions to perform one or more processes described herein. Thus, embodiments described herein are not limited to any specific combination of hardware circuitry and software. The term “configured to,” as used herein, may refer to an arrangement of software, device(s), and / or hardware for performing and / or enabling one or more functions (e.g., actions, processes, steps of a process, and / or the like). For example, “a processor configured to” may refer to a processor that executes software instructions (e.g., program code) that cause the processor to perform one or more functions.

[0194] Referring now to FIG. 6, shown is a graph showing performance of disclosed techniques for improved model training using reinforcement learning and feature dropout compared to other techniques for model training, according to some nonlimiting embodiments or aspects.

[0195] As shown in FIG. 6, the vertical axis (e.g., y-axis) is associated with a performance metric (e.g., F1 score), and the horizontal axis (e.g., x-axis) shows cutoff values. For example, cutoff may be the threshold used to determine the predicted classification based on the output probabilities of the machine learning model (e.g., when the cutoff is set to a selected threshold value, if the predicted probability satisfies (e.g., exceeds) the selected threshold value, the data record is classified as the positive class; otherwise, it is classified as the negative class). Adjusting the cutoff can affect the model's precision and recall, therefore the F1 score. Three curves are graphed for the disclosed techniques (“Feature Dropout”) 601 , hidden layer node dropout techniques (“Traditional Dropout”) 602, and a machine learning model without dropout (“Baseline”) 603. The disclosed techniques outperform both the hidden layer5Y77189.DOCX Page 55 of 69Attorney Docket No. 08223-2405237 (8270W001 ) node dropout techniques and the machine learning model without dropout across the entire range of cutoff values.

[0196] Although embodiments have been described in detail for the purpose of illustration, it is to be understood that such detail is solely for that purpose and that the disclosure is not limited to the disclosed embodiments or aspects, but, on the contrary, is intended to cover modifications and equivalent arrangements that are within the spirit and scope of the appended claims. For example, it is to be understood that the present disclosure contemplates that, to the extent possible, one or more features of any embodiment or aspect can be combined with one or more features of any other embodiment or aspect.5Y77189.DOCX Page 56 of 69

Claims

Attorney Docket No. 08223-2405237 (8270W001 )WHAT IS CLAIMED IS:1 . A system, comprising: at least one processor configured to: receive a pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; and adjust the plurality of weights of the feature selection layer of the second machine learning model based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.

2. The system of claim 1 , wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

3. The system of claim 1 , wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.5Y77189.DOCX Page 57 of 69Attorney Docket No. 08223-2405237 (8270W001 )4. The system of claim 3, wherein the at least one reinforcement learning reward comprises the feature importance reward, the loss-based reward, and the divergence-based reward.

5. The system of claim 4, wherein the feature importance reward is based on a Pearson rank correlation and a reference feature importance ranking, the loss-based reward is based on a cross-entropy loss, and the divergence-based reward is based on a Kullback-Leibler (KL) divergence.

6. The system of claim 4, wherein the feature importance reward rewards a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement learning agent for a loss, and wherein the divergence-based reward penalizes the reinforcement learning agent for a divergence.

7. The system of claim 6, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and the divergencebased reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

8. The system of claim 1 , wherein the at least one processor is further configured to: retrain the second plurality of hidden layers based on the training data and the feature selection layer having the adjusted plurality of weights to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.5Y77189.DOCX Page 58 of 69Attorney Docket No. 08223-2405237 (8270W001 )9. The system of claim 1 , wherein the at least one processor is further configured to: deploy the pretrained machine learning model in a production environment to generate outputs based on incoming data records; and detect at least one anomaly in the production environment; wherein receiving the pretrained machine learning model comprises receiving the pretrained machine learning model in a development environment based on detecting the at least one anomaly in the production environment, wherein initializing the second machine learning model comprises initializing the second machine learning model in the development environment based on receiving the pretrained machine learning model in the development environment, and wherein adjusting the plurality of weights of the feature selection layer comprises adjusting the plurality of weights of the feature selection layer based on receiving the pretrained machine learning model in the development environment.

10. The system of claim 9, wherein the at least one anomaly comprises at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.1 1 . The system of claim 9, wherein the at least one processor is further configured to: deploy the second machine learning model in the production environment.

12. A computer-implemented method, comprising: receiving, with at least one processor, a pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data5Y77189.DOCX Page 59 of 69Attorney Docket No. 08223-2405237 (8270W001 ) records, each data record of the plurality of data records comprising a plurality of features; initializing, with at least one processor, a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; and adjusting, with at least one processor, the plurality of weights of the feature selection layer of the second machine learning model based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.

13. The computer-implemented method of claim 12, wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

14. The computer-implemented method of claim 12, wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

15. The computer-implemented method of claim 14, wherein the feature importance reward is based on a Pearson rank correlation and a reference feature importance ranking, the loss-based reward is based on a cross-entropy loss; and the divergence-based reward is based on a Kullback-Leibler (KL) divergence.

16. The computer-implemented method of claim 14, wherein the feature importance reward rewards a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a5Y77189.DOCX Page 60 of 69Attorney Docket No. 08223-2405237 (8270W001 ) reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement learning agent for a loss, and wherein the divergence-based reward penalizes the reinforcement learning agent for a divergence, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and the divergence-based reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

17. The computer-implemented method of claim 12, further comprising retraining, with at least one processor, the second plurality of hidden layers based on the training data and the feature selection layer having the adjusted plurality of weights to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters.

18. The computer-implemented method of claim 12, further comprising: deploying, with at least one processor, the pretrained machine learning model in a production environment to generate outputs based on incoming data records; and detecting, with at least one processor, at least one anomaly in the production environment; wherein receiving the pretrained machine learning model comprises receiving the pretrained machine learning model in a development environment based on detecting the at least one anomaly in the production environment, wherein initializing the second machine learning model comprises initializing the second machine learning model in the development environment based on receiving the pretrained machine learning model in the development environment, and5Y77189.DOCX Page 61 of 69Attorney Docket No. 08223-2405237 (8270W001 ) wherein adjusting the plurality of weights of the feature selection layer comprises adjusting the plurality of weights of the feature selection layer based on receiving the pretrained machine learning model in the development environment.

19. The computer-implemented method of claim 18, further comprising deploying, with at least one processor, the second machine learning model in the production environment.

20. A computer program product comprising at least one non- transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to: receive a pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; and adjust the plurality of weights of the feature selection layer of the second machine learning model based on the training data and at least one reinforcement learning reward to provide an adjusted plurality of weights.21 . A system, comprising: a production environment system configured to: deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records, the pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based5Y77189.DOCX Page 62 of 69Attorney Docket No. 08223-2405237 (8270W001 ) on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; and detect at least one anomaly in the production environment; and a development environment system configured to: based on the production environment system detecting the at least one anomaly in the production environment, receive the pretrained machine learning model in a development environment; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; adjust the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights; retrain the second plurality of hidden layers based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters; and communicate the tuned machine learning model to the production environment system, wherein the production environment system is further configured to: deploy the tuned machine learning model in the production environment to generate further outputs based on further incoming data records. Y77189.DOCX Page 63 of 69Attorney Docket No. 08223-2405237 (8270W001 )22. The system of claim 21 , wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

23. The system of claim 21 or claim 22, wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

24. The system of any of claims 21 -23, wherein the at least one reinforcement learning reward comprises the feature importance reward, the lossbased reward, and divergence-based reward.

25. The system of claim 23 or claim 24, wherein the feature importance reward is based on a Pearson rank correlation and a reference feature importance ranking, the loss-based reward is based on a cross-entropy loss; and the divergence-based reward is based on a Kullback-Leibler (KL) divergence.

26. The system of any of claims 23-25, wherein the feature importance reward rewards a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement learning agent for a loss, and wherein the divergence-based reward penalizes the reinforcement learning agent for a divergence.

27. The system of any of claims 23-26, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and the divergence-based reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.5Y77189.DOCX Page 64 of 69Attorney Docket No. 08223-2405237 (8270W001 )28. The system of any of claims 21 -27, wherein the at least one anomaly comprises at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.

29. A computer-implemented method, comprising: deploying, with at least one processor, a pretrained machine learning model in a production environment to generate outputs based on incoming data records, the pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; and detecting, with at least one processor, at least one anomaly in the production environment; based on detecting the at least one anomaly in the production environment: receiving, with at least one processor, the pretrained machine learning model in a development environment; initializing, with at least one processor, a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers, the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; adjusting, with at least one processor, the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of: the training data,5Y77189.DOCX Page 65 of 69Attorney Docket No. 08223-2405237 (8270W001 ) the incoming data records, or any combination thereof to provide an adjusted plurality of weights; retraining, with at least one processor, the second plurality of hidden layers based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters; and deploying, with at least one processor, the tuned machine learning model in the production environment to generate further outputs based on further incoming data records.

30. The computer-implemented method of claim 29, wherein initializing comprises at least one of: copying parameters of the first input layer as the second input layer; copying parameters of the first plurality of hidden layers as the second plurality of hidden layers; or any combination thereof.

31. The computer-implemented method of claim 29 or claim 30, wherein the at least one reinforcement learning reward comprises at least two of: a feature importance reward, a loss-based reward, or a divergence-based reward.

32. The computer-implemented method of claim 31 , wherein the feature importance reward rewards a reinforcement learning agent when a feature importance ranking associated with the adjusted plurality of weights correlates with a reference feature importance ranking, wherein the loss-based reward penalizes the reinforcement learning agent for a loss, and wherein the divergence-based reward penalizes the reinforcement learning agent for a divergence.5Y77189.DOCX Page 66 of 69Attorney Docket No. 08223-2405237 (8270W001 )33. The computer-implemented method of claim 31 or claim 32, wherein a total reward is determined based on the feature importance reward, the loss-based reward, and divergence-based reward, and wherein adjusting the plurality of weights comprises adjusting the plurality of weights to at least one of: increase the total reward, randomly adjust at least one weight of the plurality of weights, or any combination thereof.

34. The computer-implemented method of any of claims 29-33, wherein the at least one anomaly comprises at least one of: a shift between a first distribution of the incoming data records with respect to a second distribution of the training data, a shift between a third distribution of the incoming data records for a first time period with respect to a fourth distribution of the incoming data records for a second time period, an increase in an error rate of the outputs generated by the pretrained machine learning model based on the incoming data records, or any combination thereof.

35. A computer program product comprising at least one non- transitory computer-readable medium including program instructions that, when executed by at least one processor, cause the at least one processor to: deploy a pretrained machine learning model in a production environment to generate outputs based on incoming data records, the pretrained machine learning model comprising a first input layer and a first plurality of hidden layers, the pretrained machine learning model trained based on training data comprising a plurality of data records, each data record of the plurality of data records comprising a plurality of features; and detect at least one anomaly in the production environment; based on detecting the at least one anomaly in the production environment: receive the pretrained machine learning model in a development environment; initialize a second machine learning model comprising a second input layer, a feature selection layer, and a second plurality of hidden layers, the second input layer initialized based on the first input layer, the second plurality of hidden layers initialized based on the first plurality of hidden layers,5Y77189.DOCX Page 67 of 69Attorney Docket No. 08223-2405237 (8270W001 ) the feature selection layer comprising a plurality of weights comprising a respective weight for each respective feature of the plurality of features, each respective weight of the plurality of weights associated with a respective dropout rate for the respective feature of the plurality of features; adjust the plurality of weights of the feature selection layer of the second machine learning model based on at least one reinforcement learning reward and at least one of: the training data, the incoming data records, or any combination thereof to provide an adjusted plurality of weights; retrain the second plurality of hidden layers based on the feature selection layer having the adjusted plurality of weights and at least one of: the training data, the incoming data records, or any combination thereof to provide adjusted parameters of the second plurality of hidden layers, wherein a tuned machine learning model comprises the second machine learning model having the second input layer, the feature selection layer having the adjusted plurality of weights, and the second plurality of hidden layers having the adjusted parameters; and deploy the tuned machine learning model in the production environment to generate further outputs based on further incoming data records. Y77189.DOCX Page 68 of 69

Images