System, method, and computer program product for determining attention patterns in state space models

Abstract

Systems, methods, and computer program products are provided for determining attention patterns in state space models. An example system includes at least one processor to receive input data for a state space machine learning model, generate an input sequence for the state space machine learning model based on the input data, where the input sequence includes a sequence of tokens associated with the input data, assign a plurality of weights to the sequence of tokens of the input sequence, provide the input sequence to a first block of the state space machine learning model, obtain an attention matrix based on the first block of the state space machine learning model, and display an attention pattern based on the attention matrix in a user interface.

Description

Attorney Docket No.: 08223-2506144 (9949WO01)SYSTEM, METHOD, AND COMPUTER PROGRAM PRODUCT FOR DETERMINING ATTENTION PATTERNS IN STATE SPACE MODELS CROSS REFERENCE TO RELATED APPLICATION

[0001] This application claims priority to U.S. Provisional Patent Application No. 63 / 741,571, filed January 3, 2025, the disclosure of which is hereby incorporated by reference in its entirety.BACKGROUND1. Technical Field

[0002] This disclosure relates generally to state space machine learning models and, in non-limiting embodiments or aspects, to systems, methods, and computer program products for determining attention patterns of state space machine learning models.2. Technical Considerations

[0003] State space machine learning models may refer to a class of probabilistic graphical models that describe a probabilistic dependence between a latent state variable and an observed measurement provided as an input. The latent state variable or the observed measurement can either be continuous or discrete. In some instances, state space machine learning models may include machine learning models that use state variables to describe a system by a set of first-order differential equations (e.g., rather than by one or more nth-order difference equations). State space machine learning models may provide general frameworks for analyzing deterministic and stochastic dynamical systems that are measured and / or observed through a stochastic process. Such frameworks have been successfully applied in engineering, statistics, computer science, and economics to solve a broad range of dynamical systems problems.

[0004] However, current model analysis techniques do not provide for gaining a proper understanding of the functionality of the architecture of a state space machine learning model. For example, it is unclear how the arrangement (e.g., the order) of an input sequence, for example, an input sequence of patches of an image, may affect the performance of state space machine learning models as compared to other machine learning model architectures, such as transformer machine learning models.69W0191.DOCX Page 1 of 58Attorney Docket No.: 08223-2506144 (9949WO01)SUMMARY

[0005] Accordingly, provided are improved systems, methods, and computer program products for determining attention patterns of state space machine learning models.

[0006] According to some non-limiting embodiments or aspects, provided is a system for determining attention patterns of state space machine learning models, that includes at least one processor configured to: receive input data for a state space machine learning model; generate an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data; assign a plurality of weights to the sequence of tokens of the input sequence; provide the input sequence to a first block of the state space machine learning model; obtain an attention matrix based on the input sequence at the first block of the state space machine learning model; and display an attention pattern based on the attention matrix in a user interface.

[0007] In some non-limiting embodiments or aspects, the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein, when providing the input sequence to the first block of the state space machine learning model, the at least one processor is configured to: provide the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following: a cross layout ordering method; a diagonal layout ordering method; a z-curve layout ordering method; a Hilbert layout ordering method; a Peano layout ordering method; oar spiral layout ordering method.

[0008] In some non-limiting embodiments or aspects, the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the at least one processor is further configured to: obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model; combine the first attention matrix and the second attention matrix to provide a combined attention matrix; and apply a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor69W0191.DOCX Page 2 of 58Attorney Docket No.: 08223-2506144 (9949WO01)is configured to: display an attention pattern based on the reduced dimension combined attention matrix in the user interface.

[0009] In some non-limiting embodiments or aspects, the input data comprises a first data record and a second data record, and wherein, when generating the input sequence for the state space machine learning model, the at least one processor is configured to: generate a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; and generate a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; and wherein, when assigning the plurality of weights to the sequence of tokens of the input sequence, the at least one processor is configured to: assign a first plurality of weights to the first sequence of tokens of the first input sequence; and assign a second plurality of weights to the second sequence of tokens of the second input sequence; wherein, when providing the input sequence to the first block of the state space machine learning model, the at least one processor is configured to: provide the first input sequence to the first block of the state space machine learning model; and provide the second input sequence to the first block of the state space machine learning model; wherein, when obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model, the at least one processor is configured to: obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model; and obtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model.

[0010] In some non-limiting embodiments or aspects, the at least one processor is further configured to: aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix; and apply a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; and wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to: display an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.69W0191.DOCX Page 3 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0011] In some non-limiting embodiments or aspects, the at least one processor is further configured to: adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.

[0012] In some non-limiting embodiments or aspects, the at least one processor is further configured to: receive a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to: display the attention pattern based on the attention matrix according to the display mode in the user interface.

[0013] According to some non-limiting embodiments or aspects, provided is a computer-implemented method for determining attention patterns of state space machine learning models, that includes: receiving, with at least one processor, input data for a state space machine learning model; generating, with at least one processor, an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data; assigning, with at least one processor, a plurality of weights to the sequence of tokens of the input sequence; providing, with at least one processor, the input sequence to a first block of the state space machine learning model; obtaining, with at least one processor, an attention matrix based on the input sequence at the first block of the state space machine learning model; and displaying, with at least one processor, an attention pattern based on the attention matrix in a user interface.

[0014] In some non-limiting embodiments or aspects, the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein providing the input sequence to the first block of the state space machine learning model comprises: providing the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following: a cross layout ordering method; a diagonal layout ordering method; a z-curve layout69W0191.DOCX Page 4 of 58Attorney Docket No.: 08223-2506144 (9949WO01)ordering method; a Hilbert layout ordering method; a Peano layout ordering method; oar spiral layout ordering method.

[0015] In some non-limiting embodiments or aspects, the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the method further comprises: obtaining a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model; combining the first attention matrix and the second attention matrix to provide a combined attention matrix; and applying a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises: displaying an attention pattern based on the reduced dimension combined attention matrix in the user interface.

[0016] In some non-limiting embodiments or aspects, the input data comprises a first data record and a second data record, and wherein generating the input sequence for the state space machine learning model comprises: generating a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; and generating a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; and where is assigning the plurality of weights to the sequence of tokens of the input sequence comprises: assigning a first plurality of weights to the first sequence of tokens of the first input sequence; and assigning a second plurality of weights to the second sequence of tokens of the second input sequence; wherein providing the input sequence to the first block of the state space machine learning model comprises: providing the first input sequence to the first block of the state space machine learning model; and providing the second input sequence to the first block of the state space machine learning model; wherein obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model comprises: obtaining a first attention matrix based on the first input sequence at the first block of the state space machine learning model; and obtaining a second attention matrix based on the second input sequence at the first block of the state space machine learning model.69W0191.DOCX Page 5 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0017] In some non-limiting embodiments or aspects, the method further comprises: aggregating the first attention matrix and the second attention matrix to provide an aggregated attention matrix; and applying a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises: displaying an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

[0018] In some non-limiting embodiments or aspects, the method further comprises: adjusting one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.

[0019] In some non-limiting embodiments or aspects, the method further comprises: receiving a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises: displaying the attention pattern based on the attention matrix according to the display mode in the user interface.

[0020] According to some non-limiting embodiments or aspects, provided is a computer program product for determining attention patterns of state space machine learning models, that includes 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 input data for a state space machine learning model; generate an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data; assign a plurality of weights to the sequence of tokens of the input sequence; provide the input sequence to a first block of the state space machine learning model; obtain an attention matrix based on the input sequence at the first block of the state space machine learning model; and display an attention pattern based on the attention matrix in a user interface.

[0021] In non-limiting embodiments or aspects, the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein, the program instructions that cause the at least one69W0191.DOCX Page 6 of 58Attorney Docket No.: 08223-2506144 (9949WO01)processor to provide the input sequence to the first block of the state space machine learning model, cause the at least one processor to: provide the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following: a cross layout ordering method; a diagonal layout ordering method; a z-curve layout ordering method; a Hilbert layout ordering method; a Peano layout ordering method; oar spiral layout ordering method.

[0022] In non-limiting embodiments or aspects, the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the program instructions further cause the at least one processor to: obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model; combine the first attention matrix and the second attention matrix to provide a combined attention matrix; and apply a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; and wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, cause the at least one processor to: display an attention pattern based on the reduced dimension combined attention matrix in the user interface.

[0023] In non-limiting embodiments or aspects, the input data comprises a first data record and a second data record, and wherein, the program instructions that cause the at least one processor to generate the input sequence for the state space machine learning model, cause the at least one processor to: generate a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; generate a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; and wherein, the program instructions that cause the at least one processor to assign the plurality of weights to the sequence of tokens of the input sequence, cause the at least one processor to: assign a first plurality of weights to the first sequence of tokens of the first input sequence; and assign a second plurality of weights to the second sequence of tokens of the second input sequence; wherein, the69W0191.DOCX Page 7 of 58Attorney Docket No.: 08223-2506144 (9949WO01)program instructions that cause the at least one processor to provide the input sequence to the first block of the state space machine learning model, cause the at least one processor to: provide the first input sequence to the first block of the state space machine learning model; and provide the second input sequence to the first block of the state space machine learning model; wherein, the program instructions that cause the at least one processor to obtain the attention matrix based on the input sequence at the first block of the state space machine learning model, cause the at least one processor to: obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model; and obtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model; aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix; and apply a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; and wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, cause the at least one processor to: display an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

[0024] In non-limiting embodiments or aspects, the program instructions further cause at least one processor to: adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.

[0025] In non-limiting embodiments or aspects, the program instructions further cause the at least one processor to: receive a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to: display the attention pattern based on the attention matrix according to the display mode in the user interface.

[0026] Other non-limiting embodiments or aspects will be set forth in the following numbered clauses:69W0191.DOCX Page 8 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0027] Clause 1: A system, comprising: at least one processor configured to: receive input data for a state space machine learning model; generate an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data; assign a plurality of weights to the sequence of tokens of the input sequence; provide the input sequence to a first block of the state space machine learning model; obtain an attention matrix based on the input sequence at the first block of the state space machine learning model; and display an attention pattern based on the attention matrix in a user interface.

[0028] Clause 2: The system of clause 1 , wherein the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein, when providing the input sequence to the first block of the state space machine learning model, the at least one processor is configured to: provide the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following: a cross layout ordering method; a diagonal layout ordering method; a z-curve layout ordering method; a Hilbert layout ordering method; a Peano layout ordering method; oar spiral layout ordering method.

[0029] Clause 3: The system of clause 1 or 2, wherein the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the at least one processor is further configured to: obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model; combine the first attention matrix and the second attention matrix to provide a combined attention matrix; and apply a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; and wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to: display an attention pattern based on the reduced dimension combined attention matrix in the user interface.

[0030] Clause 4: The system of any of clauses 1-3, wherein the input data comprises a first data record and a second data record, and wherein, when generating the input sequence for the state space machine learning model, the at least one69W0191.DOCX Page 9 of 58Attorney Docket No.: 08223-2506144 (9949WO01)processor is configured to: generate a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; and generate a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; and wherein, when assigning the plurality of weights to the sequence of tokens of the input sequence, the at least one processor is configured to: assign a first plurality of weights to the first sequence of tokens of the first input sequence; and assign a second plurality of weights to the second sequence of tokens of the second input sequence; wherein, when providing the input sequence to the first block of the state space machine learning model, the at least one processor is configured to: provide the first input sequence to the first block of the state space machine learning model; and provide the second input sequence to the first block of the state space machine learning model; wherein, when obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model, the at least one processor is configured to: obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model; and obtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model.

[0031] Clause 5: The system of any of clauses 1-4, wherein the at least one processor is further configured to: aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix; and apply a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; and wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to: display an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

[0032] Clause 6: The system of any of clauses 1-5, wherein the at least one processor is further configured to: adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.69W0191.DOCX Page 10 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0033] Clause 7: The system of any of clauses 1-6, wherein the at least one processor is further configured to: receive a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to: display the attention pattern based on the attention matrix according to the display mode in the user interface.

[0034] Clause 8: A computer-implemented method, comprising: receiving, with at least one processor, input data for a state space machine learning model; generating, with at least one processor, an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data; assigning, with at least one processor, a plurality of weights to the sequence of tokens of the input sequence; providing, with at least one processor, the input sequence to a first block of the state space machine learning model; obtaining, with at least one processor, an attention matrix based on the input sequence at the first block of the state space machine learning model; and displaying, with at least one processor, an attention pattern based on the attention matrix in a user interface.

[0035] Clause 9: The computer-implemented method of clause 8, wherein the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein providing the input sequence to the first block of the state space machine learning model comprises: providing the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following: a cross layout ordering method; a diagonal layout ordering method; a z-curve layout ordering method; a Hilbert layout ordering method; a Peano layout ordering method; oar spiral layout ordering method.

[0036] Clause 10: The computer-implemented method of clause 8 or 9, wherein the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the method further comprises: obtaining a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model; combining the first attention69W0191.DOCX Page 11 of 58Attorney Docket No.: 08223-2506144 (9949WO01)matrix and the second attention matrix to provide a combined attention matrix; and applying a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises: displaying an attention pattern based on the reduced dimension combined attention matrix in the user interface.

[0037] Clause 11: The computer-implemented method of any of clauses 8-10, wherein the input data comprises a first data record and a second data record, and wherein generating the input sequence for the state space machine learning model comprises: generating a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; and generating a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; and where is assigning the plurality of weights to the sequence of tokens of the input sequence comprises: assigning a first plurality of weights to the first sequence of tokens of the first input sequence; and assigning a second plurality of weights to the second sequence of tokens of the second input sequence; wherein providing the input sequence to the first block of the state space machine learning model comprises: providing the first input sequence to the first block of the state space machine learning model; and providing the second input sequence to the first block of the state space machine learning model; wherein obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model comprises: obtaining a first attention matrix based on the first input sequence at the first block of the state space machine learning model; and obtaining a second attention matrix based on the second input sequence at the first block of the state space machine learning model.

[0038] Clause 12: The computer-implemented method of any of clauses 8-11, wherein the method further comprises: aggregating the first attention matrix and the second attention matrix to provide an aggregated attention matrix; and applying a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises: displaying an69W0191.DOCX Page 12 of 58Attorney Docket No.: 08223-2506144 (9949WO01)attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

[0039] Clause 13: The computer-implemented method of any of clauses 8-12, wherein the method further comprises: adjusting one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.

[0040] Clause 14: The computer-implemented method of any of clauses 8-13, wherein the method further comprises: receiving a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises: displaying the attention pattern based on the attention matrix according to the display mode in the user interface.

[0041] Clause 15: 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 input data for a state space machine learning model; generate an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data; assign a plurality of weights to the sequence of tokens of the input sequence; provide the input sequence to a first block of the state space machine learning model; obtain an attention matrix based on the input sequence at the first block of the state space machine learning model; and display an attention pattern based on the attention matrix in a user interface.

[0042] Clause 16: The computer program product of clause 15, wherein the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein, the program instructions that cause the at least one processor to provide the input sequence to the first block of the state space machine learning model, cause the at least one processor to: provide the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following: a cross layout ordering method; a diagonal layout ordering method; a69W0191.DOCX Page 13 of 58Attorney Docket No.: 08223-2506144 (9949WO01)z-curve layout ordering method; a Hilbert layout ordering method; a Peano layout ordering method; oar spiral layout ordering method.

[0043] Clause 17: The computer program product of clause 15 or 16, wherein the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the program instructions further cause the at least one processor to: obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model; combine the first attention matrix and the second attention matrix to provide a combined attention matrix; and apply a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; and wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, cause the at least one processor to: display an attention pattern based on the reduced dimension combined attention matrix in the user interface.

[0044] Clause 18: The computer program product of any of clauses 15-17, wherein the input data comprises a first data record and a second data record, and wherein, the program instructions that cause the at least one processor to generate the input sequence for the state space machine learning model, cause the at least one processor to: generate a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; generate a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; and wherein, the program instructions that cause the at least one processor to assign the plurality of weights to the sequence of tokens of the input sequence, cause the at least one processor to: assign a first plurality of weights to the first sequence of tokens of the first input sequence; and assign a second plurality of weights to the second sequence of tokens of the second input sequence; wherein, the program instructions that cause the at least one processor to provide the input sequence to the first block of the state space machine learning model, cause the at least one processor to: provide the first input sequence to the first block of the state space machine learning model; and provide the second input sequence to the first block of the state space machine69W0191.DOCX Page 14 of 58Attorney Docket No.: 08223-2506144 (9949WO01)learning model; wherein, the program instructions that cause the at least one processor to obtain the attention matrix based on the input sequence at the first block of the state space machine learning model, cause the at least one processor to: obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model; and obtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model; aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix; and apply a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, cause the at least one processor to: display an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

[0045] Clause 19: The computer program product of any of clauses 15-18, wherein the program instructions further cause at least one processor to: adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.

[0046] Clause 20: The computer program product of any of clauses 15-19, wherein the program instructions further cause the at least one processor to: receive a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to: display the attention pattern based on the attention matrix according to the display mode in the user interface.

[0047] 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 of69W0191.DOCX Page 15 of 58Attorney Docket No.: 08223-2506144 (9949WO01)illustration and description only and are not intended as a definition of the limits of the disclosed subject matter.BRIEF DESCRIPTION OF THE DRAWINGS

[0048] 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:

[0049] FIG. 1 is a schematic diagram of a system for determining attention patterns of state space machine learning models, according to some non-limiting embodiments or aspects;

[0050] FIG. 2 is a flow diagram for a process for determining attention patterns of state space machine learning models, according to some non-limiting embodiments or aspects;

[0051] FIGS. 3A-3D are schematic diagrams of an exemplary implementation of a process for determining attention patterns of state space machine learning models, according to some non-limiting embodiments or aspects;

[0052] FIG. 3E is an illustration of user interfaces that display attention patterns of a state space machine learning model, according to some non-limiting embodiments or aspects;

[0053] FIG. 3F is an illustration of a user interface that displays attention patterns of a state space machine learning model, according to some non-limiting embodiments or aspects;

[0054] FIG. 3G is an illustration of a plurality of scan patterns of a state space machine learning model, according to some non-limiting embodiments or aspects;

[0055] FIG. 4 is a diagram of an exemplary environment in which systems, methods, and / or computer program products, described herein, may be implemented, according to some non-limiting embodiments or aspects; and

[0056] FIG. 5 is a schematic diagram of example components of one or more devices of FIG. 1 and / or FIG. 4, according to some non-limiting embodiments or aspects.DESCRIPTION

[0057] For purposes of the description hereinafter, the terms “end,” “upper,” “lower,” “right,” “left,” “vertical,” “horizontal,” “top,” “bottom,” “lateral,” “longitudinal,” and69W0191.DOCX Page 16 of 58Attorney Docket No.: 08223-2506144 (9949WO01)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.

[0058] 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.

[0059] 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 some 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).

[0060] 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 payment69W0191.DOCX Page 17 of 58Attorney Docket No.: 08223-2506144 (9949WO01)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.

[0061] 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.

[0062] 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 units. 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.

[0063] As used herein, the term “computing device” may refer to one or more electronic devices configured to process data. A computing device may, in some69W0191.DOCX Page 18 of 58Attorney Docket No.: 08223-2506144 (9949WO01)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, awearable 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.

[0064] 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.

[0065] 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.

[0066] 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 system69W0191.DOCX Page 19 of 58Attorney Docket No.: 08223-2506144 (9949WO01)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 configured to process online payment transactions through webpages, mobile applications, and / or the like.

[0067] 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, POS devices, mobile devices, etc.) directly or indirectly communicating in the network environment may constitute a “system.”

[0068] 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.

[0069] 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.69W0191.DOCX Page 20 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0070] Non-limiting embodiments or aspects of the disclosed subject matter are directed to systems, methods, and computer program products for determining attention patterns of state space machine learning models.

[0071] A state space machine learning model may use state variables to mathematically describe a state of a dynamic system. These state variables may represent the state of the dynamic system and track how an input is transformed into an output. In some examples, state space machine learning models are used to model time series tasks, where the state variables are time-dependent. Further, state space machine learning models have demonstrated performance on par with state of-the-art transformer models and, in some situations, the linear time complexity of state space machine learning models allows for outperforming transformers in latency critical applications.

[0072] Non-limiting embodiments of the disclosed subject matter may include a state space machine learning model that employs a selective scan mechanism. The selective scan mechanism may enable the state space machine learning model to assign different weights (e.g., trainable weights) to input tokens, which may allow for filtering out less relevant information and / or emphasize more relevant information based on the weights assigned to the input tokens. The trainable weights may function similar to an attention mechanism in a transformer machine learning model, and may be used to determine how much focus a token should place on preceding tokens.

[0073] In vision-based applications, a state space machine learning model may decompose an image into smaller patches and arrange the patches into sequences, which can then be processed by the state space machine learning model. In some non-limiting embodiments, the state space machine learning model may receive a plurality of patches as an input, which are provided sequentially. In this way, an order of the plurality of patches may be important, as each patch may collect information from preceding patches. To ensure that each patch has access to all other patches, forward and backward scans through the plurality of patches may be performed to preserve spatial locality of a set of patches within a same row and / or same column. For each patch, the model aggregates attention from a route used to scan patches to determine which patches to focus on.

[0074] In some non-limiting embodiments, a state space machine learning model may include a hierarchy of blocks, where each block may be designed to learn attention between patches that are based on an input (e.g., an input image). In some69W0191.DOCX Page 21 of 58Attorney Docket No.: 08223-2506144 (9949WO01)non-limiting embodiments, the disclosed subject matter provides for extracting information associated with the attention learned at each block of the state space machine learning model and applying dimensionality reduction (DR) techniques to reveal attention patterns across blocks of the state space machine learning model. Furthermore, the disclosed subject matter provides for profiling (e.g., storing) attention patterns at each patch to identify how attention is spatially distributed relative to a position of a patch, which may provide for hierarchical attention learning patterns within the state space machine learning model.

[0075] Further, the disclosed subject matter may include a model analysis system that is configured to receive input data for a state space machine learning model, generate an input sequence for the state space machine learning model based on the input data, where the input sequence comprises a sequence of tokens associated with the input data, assign a plurality of weights to the sequence of tokens of the input sequence, provide the input sequence to a first block of the state space machine learning model, obtain an attention matrix based on the input sequence at the first block of the state space machine learning model, and display an attention pattern based on the attention matrix in a user interface.

[0076] In some non-limiting embodiments or aspects, the sequence of tokens of the input sequence may include a plurality of patches (e.g., patches of an image), where each patch of the plurality of patches is associated with a column and a row of a grid and the grid represents the input data. When providing the input sequence to the first block of the state space machine learning model, the model analysis system may be configured to provide the input sequence to a first block of the state space machine learning model according to a patch ordering method, where the patch ordering method may include at least one of the following: a cross layout ordering method, a diagonal layout ordering method, a z-curve layout ordering method, a Hilbert layout ordering method, a Peano layout ordering method, or a spiral layout ordering method.

[0077] In some non-limiting embodiments or aspects, the first block of the state space machine learning model may be a first block of a first stage of the state space machine learning model, the attention matrix based on the first block may be a first attention matrix, and the model analysis system may be further configured to obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model, combine the first attention matrix69W0191.DOCX Page 22 of 58Attorney Docket No.: 08223-2506144 (9949WO01)and the second attention matrix to provide a combined attention matrix, and apply a DR algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix. In some non-limiting embodiments or aspects, when displaying the attention pattern based on the attention matrix in the user interface, the model analysis system may be configured to display an attention pattern based on the reduced dimension combined attention matrix in the user interface.

[0078] In some non-limiting embodiments or aspects, the input data may include a first data record and a second data record, the attention matrix based on the first block may be a first attention matrix and, when generating the input sequence for the state space machine learning model, the model analysis system may be configured to generate a first input sequence for the state space machine learning model based on the first data record of the input data, where the first input sequence comprises a first sequence of tokens associated with the first data record, and generate a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record. In some non-limiting embodiments or aspects, when assigning the plurality of weights to the sequence of tokens of the input sequence, the model analysis system may be configured to assign a first plurality of weights to the first sequence of tokens of the first input sequence and assign a second plurality of weights to the second sequence of tokens of the second input sequence. In some non-limiting embodiments or aspects, when providing the input sequence to the first block of the state space machine learning model, the model analysis system may be configured to provide the first input sequence to the first block of the state space machine learning model and provide the second input sequence to the first block of the state space machine learning model. In some non-limiting embodiments or aspects, when obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model, the model analysis system may be configured to obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model and obtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model.

[0079] In some non-limiting embodiments or aspects, the model analysis system may be further configured to aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix and apply a DR algorithm69W0191.DOCX Page 23 of 58Attorney Docket No.: 08223-2506144 (9949WO01)to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix. In some non-limiting embodiments or aspects, when displaying the attention pattern based on the attention matrix in the user interface, the model analysis system may be configured to display an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

[0080] In some non-limiting embodiments or aspects, the model analysis system may be further configured to adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, where each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the plurality of tokens.

[0081] In some non-limiting embodiments or aspects, the model analysis system may be further configured to receive a selection of a display mode for displaying the attention pattern in the user interface, where the display mode may include a scatterplot view mode. In some non-limiting embodiments or aspects, when displaying the attention pattern based on the attention matrix in the user interface, the model analysis system may be configured to display the attention pattern based on the attention matrix according to the display mode in the user interface.

[0082] In this way, the disclosed subject matter may provide for the ability to understand the functionality of the architecture of a state space machine learning model. For example, the disclosed subject matter may provide for the ability to visually observe how the arrangement of an input sequence may affect the performance of state space machine learning models as compared to other machine learning model architectures, such as transformer machine learning models. Also, the disclosed subject matter is unique and unconventional.

[0083] Referring now to FIG. 1, shown is system 100 for determining attention patterns of state space machine learning models, according to some non-limiting embodiments or aspects. For example, system 100 may include model analysis system 102, machine learning (ML) model management database 104, user device 106, and communication network 108.

[0084] Model analysis system 102 may include one or more devices capable of receiving information from and / or communicating information to ML model management database 104 and / or user device 106 (e.g., directly via wired or wireless communication connection, indirectly via communication network 108, and / or the like).69W0191.DOCX Page 24 of 58Attorney Docket No.: 08223-2506144 (9949WO01)For example, model analysis 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, model analysis system 102 may be in communication with a data storage device (e.g., ML model management database 104), which may be local or remote to model analysis system 102. In some non-limiting embodiments or aspects, model analysis 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., ML model management database 104).

[0085] ML model management database 104 may include one or more devices capable of receiving information from and / or communicating information to model analysis system 102 and / or user device 106 (e.g., directly via wired or wireless communication connection, indirectly via communication network 108, and / or the like). For example, ML model management database 104 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, ML model management database 104 may include a data storage device. In some non-limiting embodiments or aspects, ML model management database 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, ML model management database 104 may be part of model analysis system 102 and / or part of the same system as model analysis system 102.

[0086] User device 106 may include one or more devices capable of receiving information from and / or communicating information to model analysis system 102 and / or ML model management database 104 (e.g., directly via wired or wireless communication connection, indirectly via communication network 108, and / or the like). For example, user device 106 may include a computing device, such as a mobile device, a portable computer, a desktop computer, and / or other like devices. Additionally or alternatively, each user device 106 may include a device capable of receiving information from and / or communicating information to other user devices 106 (e.g., directly via wired or wireless communication connection, indirectly via communication network 108, and / or the like). In some non-limiting embodiments or aspects, user device 106 may be part of model analysis system 102 and / or part of the69W0191.DOCX Page 25 of 58Attorney Docket No.: 08223-2506144 (9949WO01)same system as model analysis system 102. For example, model analysis system 102, ML model management database 104, and user device 106 may all be (and / or be part of) a single system and / or a single computing device.

[0087] Communication network 108 may include one or more wired and / or wireless networks. For example, communication network 108 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), a 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.

[0088] 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, 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.

[0089] Referring now to FIG. 2, shown is a flow diagram for process 200 for determining attention patterns of state space machine learning models, according to some non-limiting embodiments or aspects. The steps shown in FIG. 2 are for example purposes only. It will be appreciated that additional, fewer, different, and / or different order of steps may be used in 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 process 200 may be performed (e.g., completely, partially, and / or the like) by model analysis system 102 (e.g., at least one computing device of model analysis system 102). In some non-limiting embodiments or aspects, one or more of the steps of process 200 may be performed69W0191.DOCX Page 26 of 58Attorney Docket No.: 08223-2506144 (9949WO01)(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 model analysis system 102, such as ML model management database 104, user device 106, and / or the like.

[0090] As shown in FIG. 2, at step 202, process 200 may include receiving input data for a state space machine learning model. For example, model analysis system 102 may receiving input data for a state space machine learning model. In some nonlimiting embodiments or aspects, the input data may include a plurality of data records, such as a first data record, a second data record, a third data record, and / or the like. In some non-limiting embodiments or aspects, each data record of the plurality of data records may be associated with an image. In some non-limiting embodiments or aspects, the state space machine learning model may include a plurality of stages and each stage may include a plurality of blocks. For example, a stage may include 2 blocks, 4 blocks, 8 blocks, and / or the like.

[0091] As shown in FIG. 2, at step 204, process 200 may include generating an input sequence for the state space machine learning model based on the input data. For example, model analysis system 102 may generate the input sequence for the state space machine learning model based on the input data. In some non-limiting embodiments or aspects, the input sequence may include a sequence of tokens associated with the input data. For example, the input sequence may include a sequence of tokens associated with a data record. In some non-limiting embodiments or aspects, the sequence of tokens may be associated with patches of an image.

[0092] In some non-limiting embodiments or aspects, model analysis system 102 may generate a plurality of input sequences for the state space machine learning model based on the input data. For example, the input data may include a first data record and a second data record. In such an example, model analysis system 102 may generate a first input sequence for the state space machine learning model based on the first data record of the input data, where the first input sequence comprises a first sequence of tokens associated with the first data record, and generate a second input sequence for the state space machine learning model based on the second data record of the input data, where the second input sequence comprises a second sequence of tokens associated with the second data record.

[0093] As shown in FIG. 2, at step 206, process 200 may include assigning a plurality of weights to a sequence of tokens of the input sequence. For example, model69W0191.DOCX Page 27 of 58Attorney Docket No.: 08223-2506144 (9949WO01)analysis system 102 may assign the plurality of weights to the sequence of tokens of the input sequence. In some non-limiting embodiments or aspects, model analysis system 102 may adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence. For example, model analysis system 102 may adjust one or more weights of the plurality of weights during a training procedure (e.g., of the state space machine learning model). In some non-limiting embodiments or aspects, each weight of the plurality of weights may be associated with an indication of importance of a token of the sequence of tokens relative to other tokens of the plurality of tokens.

[0094] In some non-limiting embodiments or aspects, model analysis system 102 may assign a plurality of weights to a plurality of sequences of tokens of a plurality of input sequences. For example, model analysis system 102 may assign a first plurality of weights to the first sequence of tokens of the first input sequence and assign a second plurality of weights to the second sequence of tokens of the second input sequence.

[0095] As shown in FIG. 2, at step 208, process 200 may include obtaining an attention matrix based on a first block of the state space machine learning model. For example, model analysis system 102 may obtain the attention matrix based on the first block of the state space machine learning model. In some non-limiting embodiments or aspects, model analysis system 102 may provide the input sequence to the first block of the state space machine learning model and may obtain the attention matrix based on providing the input sequence to the first block.

[0096] In some non-limiting embodiments or aspects, the sequence of tokens of the input sequence comprises a plurality of patches, and each patch of the plurality of patches may be associated with a column and a row of a grid. The grid may represent the input data, such as a data record that includes an image. In some non-limiting embodiments or aspects, when providing the input sequence to the first block of the state space machine learning model, model analysis system 102 may provide the input sequence to the first block of the state space machine learning model according to a patch ordering method. In some non-limiting embodiments or aspects, the patch ordering method comprises at least one of the following: a cross layout ordering method, a diagonal layout ordering method, a z-curve layout ordering method, a Hilbert layout ordering method, a Peano layout ordering method, a spiral layout ordering method, or any combination thereof.69W0191.DOCX Page 28 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0097] In some non-limiting embodiments or aspects, model analysis system 102 may provide a plurality of input sequences to a first block of the state space machine learning model (e.g., according to a patch ordering method). For example, model analysis system 102 may provide a first input sequence to the first block of the state space machine learning model and provide a second input sequence to the first block of the state space machine learning model. In some non-limiting embodiments or aspects, when obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model, model analysis system 102 may obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model and obtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model. In some non-limiting embodiments or aspects, model analysis system 102 may aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix and apply a DR algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix.

[0098] In some non-limiting embodiments or aspects, the first block of the state space machine learning model may be a first block of a first stage of the state space machine learning model. In some non-limiting embodiments or aspects, model analysis system 102 may obtain a plurality of attention matrices based on the first block of the state space machine learning model. For example, model analysis system 102 may obtain a first attention matrix based on the input sequence at the first block of the first stage of the state space machine learning model and obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model. In some non-limiting embodiments or aspects, model analysis system 102 may combine the first attention matrix and the second attention matrix to provide a combined attention matrix and apply a DR algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix.

[0099] As shown in FIG. 2, at step 210, process 200 may include displaying an attention pattern based on the attention matrix. For example, model analysis system 102 may display the attention pattern based on the attention matrix in a user interface (e.g., a user interface of user device 106). In some non-limiting embodiments or aspects, model analysis system 102 may display an attention pattern based on a reduced dimension combined attention matrix in the user interface. Additionally or69W0191.DOCX Page 29 of 58Attorney Docket No.: 08223-2506144 (9949WO01)alternatively, model analysis system 102 may display an attention pattern based on a reduced dimension aggregated attention matrix in the user interface.

[0100] In some non-limiting embodiments or aspects, model analysis system 102 may receive a selection (e.g., a selection made via a user element of a user interface) of a display mode for displaying the attention pattern in the user interface, where the display mode may include a plurality of view modes. In some non-limiting embodiments or aspects, the plurality of view modes may include a scatterplot view mode and a patch view mode. In some non-limiting embodiments or aspects, when displaying the attention pattern based on the attention matrix in a user interface, model analysis system 102 may display an attention pattern based on an attention matrix, according to the display mode in the user interface.

[0101] Referring now to FIGS. 3A-3D, shown are schematic diagrams of exemplary implementations 300 of a process (e.g., process 200) for determining attention patterns of state space machine learning models, according to some non-limiting embodiments or aspects. In some non-limiting embodiments or aspects, one or more of the steps of the process may be performed (e.g., completely, partially, etc.) by model analysis system 102 (e.g., one or more devices of model analysis system 102). In some non-limiting embodiments or aspects, one or more of the steps of the process may be performed (e.g., completely, partially, etc.) by another device or a group of devices separate from or including model analysis system 102 (e.g., one or more devices of model analysis system 102), ML model management database 104, and / or user device 106.

[0102] As shown by reference number 302 in FIG. 3A, model analysis system 102 may receive input data for a state space machine learning model. In some non-limiting embodiments, the input data may include an array of input images, such as an array of red-green-blue (RGB) input images. As further shown in FIG. 3A, the state space machine learning model may include a plurality of stages, where each stage comprises a plurality of blocks (e.g., a plurality of scan-merge blocks) along with a plurality of sampling operations (e.g., downsampling operations). In some non-limiting embodiments or aspects, one or more stages of the plurality of stages may be the same as or similar to another stage. In some non-limiting embodiments or aspects, one or more stages of the plurality of stages may be different from another stage. Stacking scan-merge blocks may be used to form a stage and, in the example of FIG. 3A, four such stages are shown. Between stages, latent representations may be69W0191.DOCX Page 30 of 58Attorney Docket No.: 08223-2506144 (9949WO01)downsampled to extract features (e.g., in a fashion similar to a convolutional neural network). Finally, through a layer (e.g., an output layer that includes a fully connected layer), the latent representations are transformed into a vector (e.g., an output vector), with the length of the vector corresponding to a number of possible classifications. Each element of the vector may represent a probability that an input corresponds to a particular classification. In some non-limiting embodiments or aspects, a final stage (e.g., shown as “stage 3” in FIG. 3A) may include a pooling layer (e.g., a layer that performs a pooling operation, such as a linear average pooling operation). In some non-limiting embodiments or aspects, the output of the state space machine learning model may include a probability of a classification of an input (e.g., an input image).

[0103] As shown by reference number 304 in FIG. 3B, model analysis system 102 may generate an input sequence for the state space machine learning model based on the input data. In some non-limiting embodiments, each input image may have equal width and height, denoted as “s.” In some examples, the array of input images has a size of s x s x 3. In some non-limiting embodiments or aspects, model analysis system 102 may decompose an input image into p x p patches, each of size pz x pz, such that s = p x pz. In some non-limiting embodiments, model analysis system 102 may configure each patch as a token, and the p x p patches may be arranged into an input sequence (e.g., arranged into a sequence while maintaining the spatial locality between patches). For example, model analysis system 102 may configure each patch as a token and the plurality of tokens corresponding to the plurality of patches for an input (e.g., an input image) may be arranged into an input sequence based on a scan pattern for the patches of an input image. The input sequence may be treated similarly to textual data and provided as an input (e.g., provided sequentially as an input) to the state space machine learning model (e.g., for learning). In some non-limiting embodiments, model analysis system 102 may generate an input sequence for the state space machine learning model for each scan pattern of a plurality of scan patterns (e.g., the four scan patterns shown in FIG. 3A).

[0104] In some non-limiting embodiments or aspects, model analysis system 102 may perform a plurality of different scan patterns for processing the patches in a block of a stage of the state space machine learning model. As further shown in FIG. 3B, model analysis system 102 may perform four different scan patterns for processing the patches: (1) left-right and top-down, (2) top-down then left-right, (3) right-left then bottom-up, and (4) bottom-up then right-left. Each scan pattern may provide a different69W0191.DOCX Page 31 of 58Attorney Docket No.: 08223-2506144 (9949WO01)representation of dependency between patches of an input and in this way, the attention that is learned for the state space machine learning model between patches may also vary. In some non-limiting embodiments, model analysis system 102 may update state variables of the state space machine learning model to mathematically describe the state of the state space machine learning model based on the attention that is learned between blocks in a stage of the state space machine learning model and / or the attention that is learned between patches of an input.

[0105] In some non-limiting embodiments, model analysis system 102 may perform a merge operation based on each patch of an input (e.g., an input image) according to a scan pattern used on a plurality of patches of the input. In some non-limiting embodiments or aspects, the merge operation may aggregate the learned attention (e.g., a weight) for each patch across a plurality of scan patterns (e.g., the four different scan patterns shown in FIG. 3A).

[0106] As shown by reference number 306 in FIG. 3C, model analysis system 102 may assign a plurality of weights to a sequence of tokens of the input sequence. In some non-limiting embodiments or aspects, model analysis system 102 may generate a plurality of weights for the sequence of tokens of the input sequence based on a scan pattern for each input (e.g., each input image of an array of input images). For example, model analysis system 102 may input each token of the sequence of tokens to one or more blocks (e.g., all blocks of all stages, each block of a stage, a plurality of blocks of a stage, etc.) of the state space machine learning model to generate the plurality of weight associated with the sequence of tokens. The weight associated with the sequence of tokens provides an indication of an amount of focus that should be placed on a token based on one or more preceding tokens. In some non-limiting embodiments or aspects, model analysis system 102 may generate a plurality of weights for each sequence of tokens of a plurality of sequences of tokens, which are generated based on each scan pattern of a plurality of scan patterns for each input (e.g., each input image of an array of input images). In some non-limiting embodiments or aspects, model analysis system 102 may assign the plurality of weights to the sequence of tokens of the input sequence based on generating the plurality of weights associated with the sequence of tokens.

[0107] As further shown by reference number 308 in FIG. 3C, model analysis system 102 may obtain an attention matrix based on a first block of the state space machine learning model. For example, model analysis system 102 may obtain the69W0191.DOCX Page 32 of 58Attorney Docket No.: 08223-2506144 (9949WO01)attention matrix based on a first block of the state space machine learning model by generating a plurality of weights associated with each sequence of tokens for each input provided as an input to the first block of the state space machine learning model. In some non-limiting embodiments or aspects, the attention matrix may include a p2x p2attention matrix, where each element at position ( / , j) in the matrix reflects an attention strength (e.g., a weight, an attention weight, etc.) between patch / and patch j. In some non-limiting embodiments or aspects, model analysis system 102 may obtain an attention matrix based on each block of the state space machine learning model. For example, model analysis system 102 may obtain an attention matrix for each block of each stage of the state space machine learning model.

[0108] As further shown by reference number 308 in FIG. 3C, model analysis system 102 may obtain an attention matrix based on a first block of the state space machine learning model. For example, model analysis system 102 may obtain an attention matrix based on a first block of the state space machine learning model.

[0109] In some non-limiting embodiments or aspects, model analysis system 102 may generate an attention pattern. In some non-limiting embodiments or aspects, model analysis system 102 may provide a plurality of inputs, for example, a plurality of n input images, as an input to the state space machine learning model, and model analysis system 102 may collect the attention matrices from m blocks of the state space machine learning model, mx n attention matrices may be generated. In some non-limiting embodiments, model analysis system 102 may use the m x n attention matrices to generate attention patterns of the state space machine learning model.

[0110] In some non-limiting embodiments or aspects, model analysis system 102 may generate an attention pattern among blocks of the same stage of the state space machine learning model. For example, when inputting n input images into the state space machine learning model, model analysis system 102 may generate an n x (p2x p2) attention matrix. In such an example, each row may represent an attention strength (e.g., a weight) between the p2x p2patches of one input image. Further, assuming that a stage consists of m blocks, model analysis system 102 may stack m such attention matrices together, resulting in an attention matrix having a shape of (m x n) x (p2x p2). In some non-limiting embodiments or aspects, the size of an attention matrix within a stage may remain constant across blocks in that stage. In some non-limiting embodiments, model analysis system 102 may use one or more dimensionality reduction (DR) algorithms to reduce the dimensionality of the (m x n) x (p2x p2)69W0191.DOCX Page 33 of 58Attorney Docket No.: 08223-2506144 (9949WO01)attention matrix to 2D, resulting in an attention matrix having a shape of (m x n) x 2. In some non-limiting embodiments or aspects, model analysis system 102 may display a scatterplot to provide a visualization of the (m x n) x 2 attention matrix, where the values along the m-dimension may be used for coloring, and the two values along the last dimension may be used to position the points. Based on the clustering of the points in the attention pattern, model analysis system 102 may determine whether the blocks within the same stage exhibit similar attention patterns.

[0111] Additionally or alternatively, model analysis system 102 may generate an attention pattern within the same block of the state space machine learning model. In some non-limiting embodiments or aspects, the attention matrix from a single block for a single input image has a shape of (p2x p2). To augment the attention patterns and extract content-irrelevant patterns, model analysis system 102 may generate this attention matrix for n input images (e.g., 1,000 input images) and aggregate the resulting n x (p2x p2) attention matrix for possible attention pattern analysis. In some non-limiting embodiments or aspects, model analysis system 102 may model analysis system 102 may may input the n input images into the state space machine learning model to obtain the attention matrix from a targeted block of the state space machine learning model, where a shape of the attention matrix is n x (p2x p2). To extract the shared attention patterns among the input images for the targeted block, model analysis system 102 may aggregate the n x (p2x p2) attention matrix by averaging along a first dimension, resulting in a p2x p2matrix. In some non-limiting embodiments or aspects, each element at position ( / , j) of the p2x p2attention matrix represents an average amount of attention from patch / to patch j over all n input images. In some non-limiting embodiments or aspects, model analysis system 102 may use one or more DR algorithms to reduce the p2x p2attention matrix to a p2x 2 attention matrix, where the p2rows represent the p2patches. In some non-limiting embodiments or aspects, model analysis system 102 may display a scatterplot to provide a visualization of the p22D points. In some non-limiting embodiments or aspects, the color and size of each point may correspond to the row and column of a corresponding patch. Based on the clustering of the points in the attention pattern, model analysis system 102 may identify patches that share similar attention patterns.

[0112] As shown by reference number 310 in FIG. 3D, model analysis system 102 may display an attention pattern based on the attention matrix. As further shown in FIG. 3D, in the architecture of the state space machine learning model depicted in69W0191.DOCX Page 34 of 58Attorney Docket No.: 08223-2506144 (9949WO01)implementation 300 shown in FIGS. 3A-3D, there are four stages, with 2, 2, 8, and 2 blocks per stage, respectively. FIG. 3D provides an illustration of the DR results for the four stages, where each individual illustration represents the output of a different stage. The color in each figure corresponds to a block identification (which are also labeled in FIG. 3D), and distinct clusters of points are visible, with different colors separating the blocks. The illustration of FIG. 3D indicates that different blocks of a stage may exhibit different attention patterns.

[0113] Referring now to FIG. 3E, FIG. 3E illustrates user interfaces of Scatterplot view 350 and Patch view 360 that display attention patterns of a state space machine learning model, according to some non-limiting embodiments. As shown in FIG. 3E, model analysis system 102 may display an attention pattern in a user interface, according to Scatterplot view 350 and / or Patch view 360. In some non-limiting embodiments or aspects, in a first mode, model analysis system 102 may display Scatterplot view 350 that allows for selection of a stage and / or a block (e.g., from a header of Scatterplot view 350). If the block is not specified, model analysis system 102 may display mx n points based on an output of the state space machine learning model, where m represents a number of blocks in a selected stage, and n is a number of input images included in input data for the state space machine learning model. In some non-limiting embodiments, each point shown in a Scatterplot view may correspond to a p2x p2attention matrix for an input image from a selected block.

[0114] In some non-limiting embodiments or aspects, in a second mode, when a block identification is specified (e.g., via a user selection), Scatterplot view 350 will display p2points based on an output of the state space machine learning model. Each point may represent an averaged attention (e.g., an average weight) across n input images for a specific patch position. The color and size of each point may represent the column and row positions of the corresponding patch, respectively.

[0115] In some non-limiting embodiments, Patch view 360 may provide two visualization modes, which allow for visualization of patch-level details from a selected stage of a state space machine learning model. As shown by cluster patterns in FIG.3E, patches from the same row (e.g., points in Scatterplot view 350 with the same size) or the same column (e.g., points in Scatterplot view 350 with the same color) may exhibit similar attention patterns. For example, clusters labeled as c and d, reveal further clusters that share similar attention patterns with those in clusters labeled a and b, respectively. For both the first mode and the second mode, model analysis69W0191.DOCX Page 35 of 58Attorney Docket No.: 08223-2506144 (9949WO01)system 102 may use one or more DR techniques. For example, model analysis system 102 may use PCA, t-SNE, and / or LIMAP. In some non-limiting embodiments, the Scatterplot view may support zooming in and out, enabling examination of cluster details at various levels of granularity. In the second mode, a group of points may be selected and the corresponding patches for the selected points and a visualization may be provided in Patch view 360.

[0116] In a first mode, model analysis system 102 may display patches from a selected stage as gray squares. For example, as shown in FIG. 3E, the 28x28 patches from stage 1 may be visualized as 28x28 squares. Meanwhile, the selected patches from the Scatterplot view 350 may be highlighted as red squares in Patch view 360, visually indicating their spatial location within an input image. The highlighting directly corresponds to the cluster patterns displayed in Scatterplot view 360, which helps reveal the spatial relationships between patches that exhibit similar attention patterns. This coordinated visualization may provide a clearer understanding of how attention patterns are distributed across different regions of an input image. In a second mode, when a square (e.g., a patch) is selected in Patch view 360, all squares will be colored based on the attention (e.g., the weight) to the selected square. Attention values from small to large may be mapped to colors, for example, from light-yellow to dark-red.

[0117] Referring now to FIG. 3F, FIG. 3F illustrates a user interface of Patch view 370 that displays attention patterns of a state space machine learning model, according to some non-limiting embodiments. As shown in display a of FIG. 3F, a selected patch is located at row 5, column 4, and the selected patch shows a strong attention to: (1 ) itself, (2) the patches in the same row to its left, and (3) the patches to its right. In this way, it may be shown that patches in the same column with the selected patch have similar attention patterns. Further, as shown in display b and display c of FIG. 3F, these two patches, which are in the same column as the patch in display a, exhibit similar attention patterns.

[0118] Referring now to FIG. 3G, FIG. 3G illustrates a plurality of scan patterns of a state space machine learning model, according to some non-limiting embodiments. As shown in FIG. 3G, Diagonal may provide a scan pattern that results in a patch ordering that is along a diagonal of an input image, where neighboring patches maintain spatial locality but are arranged along the diagonal axis. As further shown in FIG. 3G, Morton may provide a scan pattern that results in a patch ordering based on the Morton curve (e.g., z- curve), which may refer to a space-filling curve that is69W0191.DOCX Page 36 of 58Attorney Docket No.: 08223-2506144 (9949WO01)effective at preserving spatial locality. As further shown in FIG. 3G, Spiral may provide a scan pattern that results in a patch ordering according to a spiral layout, where the innermost patch may maintain the strongest spatial locality.

[0119] Referring now to FIG. 4, depicted is a diagram of example payment processing network 400, according to non-limiting embodiments or aspects. In some non-limiting embodiments or aspects, payment processing network 400 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 400.

[0120] As shown in FIG. 4, payment processing network 400 may include transaction processing system 402, payment gateway system 412, merchant system 408, issuer system 404, acquirer system 410, and / or customer device 406. In some non-limiting embodiments or aspects, each of model analysis system 102, database 104, and / or user device 106 of FIG. 1 may be implemented by (e.g., part of) transaction processing system 402. In some non-limiting embodiments or aspects, at least one of model analysis system 102, database 104, and / or user device 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 402, such as merchant system 408, issuer system 404, acquirer system 410, customer device 406, and / or the like. For example, model analysis system 102 may be implemented by (e.g., part of) at least one of payment gateway system 412, merchant system 408, issuer system 404, acquirer system 410, and / or customer device 406. The systems and / or devices of FIG. 4 may communicate via communication network 414, which may include one or more wired and / or wireless communication networks.

[0121] Transaction processing system 402 may include one or more devices capable of receiving information from and / or communicating information to payment gateway system 412, merchant system 408, issuer system 404, acquirer system 410, customer device 406, 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. 4, transaction processing system 402 may be in communication with one or more issuer systems (e.g., issuer system 404), one or more acquirer systems (e.g., acquirer system 410), and / or one or more payment gateway systems (e.g., payment gateway system 412). Although only a single issuer system 404, a single acquirer69W0191.DOCX Page 37 of 58Attorney Docket No.: 08223-2506144 (9949WO01)system 410, and a single payment gateway system 412 are shown, it will be appreciated that transaction processing system 402 may be in communication with a plurality of issuer systems, a plurality of acquirer systems, and / or a plurality of payment gateway systems. In some non-limiting embodiments or aspects, transaction processing system 402 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 402 may be in communication with a data storage device, which may be local or remote to transaction processing system 402. In some non-limiting embodiments or aspects, transaction processing system 402 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 402 may be associated with a transaction service provider, as described herein. In some non-limiting embodiments or aspects, transaction processing system 402 may also operate as an issuer system, such that both transaction processing system 402 and issuer system 404 are a single system and / or are controlled by a single entity.

[0122] Payment gateway system 412 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 402, merchant system 408, issuer system 404, acquirer system 410, customer device 406, 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. 4, payment gateway system 412 may be in communication with one or more merchant systems (e.g., merchant system 408), one or more acquirer systems (e.g., acquirer system 410), and / or one or more transaction processing systems (e.g., transaction processing system 402). Although only a single merchant system 408, a single acquirer system 410, and a single transaction processing system 402 are shown, it will be appreciated that payment gateway system 412 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 412 may include a computing device, such as a server, a group of servers, and / or other like devices. In some nonlimiting embodiments or aspects, payment gateway system 412 may be associated with a payment gateway, as described herein.69W0191.DOCX Page 38 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0123] Merchant system 408 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 402, payment gateway system 412, issuer system 404, acquirer system 410, customer device 406, 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.4, merchant system 408 may be in communication with one or more payment gateway systems (e.g., payment gateway system 412), one or more acquirer systems (e.g., acquirer system 410), and / or one or more consumer devices (e.g., customer device 406). Although only a single payment gateway system 412, a single acquirer system 410, and a single customer device 406 are shown, it will be appreciated that merchant system 408 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 nonlimiting embodiments or aspects, merchant system 408 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 408 may be associated with a merchant, as described herein. In some nonlimiting embodiments or aspects, merchant system 408 may include a device capable of receiving information from and / or communicating information to customer device 406 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 customer device 406 and / or the like. In some non-limiting embodiments or aspects, merchant system 408 may include one or more client devices. For example, merchant system 408 may include a client device that allows a merchant to communicate information to transaction processing system 402 (e.g., via at least one of acquirer system 410 and / or payment gateway system 412). In some non-limiting embodiments or aspects, merchant system 408 (e.g., a client device thereof, a POS device thereof, and / or the like) may also operate as a payment gateway system, such that both merchant system 408 and payment gateway system 412 are a single system and / or controlled by a single entity.

[0124] Issuer system 404 may include one or more devices capable of receiving information and / or communicating information to transaction processing system 402, payment gateway system 412, merchant system 408, acquirer system 410, customer69W0191.DOCX Page 39 of 58Attorney Docket No.: 08223-2506144 (9949WO01)device 406, 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.4, issuer system 404 may be in communication with one or more transaction processing systems (e.g., transaction processing system 402) and / or one or more consumer devices (e.g., customer device 406). Although only a single transaction processing system 402 and a single customer device 406 are shown, it will be appreciated that issuer system 404 may be in communication with a plurality of transaction processing systems and / or a plurality of customer devices 406. In some non-limiting embodiments or aspects, issuer system 404 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 404 may be associated with an issuer institution, as described herein. For example, issuer system 404 may be associated with an issuer institution that issued a credit account, debit account, credit card, debit card, a payment device, and / or the like to a user associated with customer device 406.

[0125] Acquirer system 410 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 402, payment gateway system 412, merchant system 408, issuer system 404, customer device 406, 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. 4, acquirer system 410 may be in communication with one or more transaction processing systems (e.g., transaction processing system 402), one or more payment gateway systems (e.g., payment gateway system 412), and / or one or more merchant systems (e.g., merchant system 408). Although only a single transaction processing system 402, a single payment gateway system 412, and a single merchant system 408 are shown, it will be appreciated that acquirer system 410 may be in 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 410 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 410 may be associated with an acquirer institution, as described herein.

[0126] Customer device 406 may include one or more devices capable of receiving information from and / or communicating information to transaction processing system 402, payment gateway system 412, merchant system 408, issuer system 404, acquirer69W0191.DOCX Page 40 of 58Attorney Docket No.: 08223-2506144 (9949WO01)system 410, 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.4, customer device 406 may be in communication with one or more merchant systems (e.g., merchant system 408) and / or one or more issuer systems (e.g., issuer system 404). Although only a single merchant system 408 and a single issuer system 404 are shown, it will be appreciated that customer device 406 may be in communication with a plurality of merchant systems and / or a plurality of issuer systems. In some nonlimiting embodiments or aspects, customer device 406 may be associated with a user to whom a credit account, debit account, credit card, debit card, a payment device, and / or the like has been issued. In some non-limiting embodiments or aspects, customer device 406 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, customer device 406 may include a payment device, as described herein. In some non-limiting embodiments or aspects, customer device 406 may include a device capable of receiving information from and / or communicating information to other customer devices 406 (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, customer device 406 may include a device capable of receiving information from and / or communicating information to merchant system 408 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 merchant system 408 and / or the like. In some non-limiting embodiments or aspects, customer device 406 may include a client device.

[0127] In some non-limiting embodiments or aspects, transaction processing system 402 may communicate with merchant system 408 directly (e.g., via a public and / or private communication network connection and / or the like). Additionally or alternatively, transaction processing system 402 may communicate with merchant system 408 through payment gateway system 412 and / or acquirer system 410. In some non-limiting embodiments or aspects, acquirer system 410 associated with merchant system 408 may operate as payment gateway system 412 to facilitate the communication of transaction messages (e.g., authorization requests) from merchant69W0191.DOCX Page 41 of 58Attorney Docket No.: 08223-2506144 (9949WO01)system 408 to transaction processing system 402. In some non-limiting embodiments or aspects, merchant system 408 may communicate with payment gateway system 412 directly (e.g., via a public and / or private communication network connection and / or the like). For example, merchant system 408, that includes a physical POS device, may communicate with payment gateway system 412 through a public or private network to conduct card-present transactions. As another example, merchant system 408, that includes a server (e.g., a web server), may communicate with payment gateway system 412 through a public or private network, such as the Internet, to conduct card-not-present transactions.

[0128] 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 customer device 406 and / or the like) and / or transaction data associated with the transaction. For example, merchant system 408 (e.g., a client device of merchant system 408, a POS device of merchant system 408, 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 408 may communicate the authorization request to payment gateway system 412 and / or acquirer system 410. In some non-limiting embodiments or aspects, payment gateway system 412 may communicate the authorization request to acquirer system 410 and / or transaction processing system 402. Additionally or alternatively, acquirer system 410 (and / or payment gateway system 412) may communicate the authorization request to transaction processing system 402. After receiving the authorization request from merchant system 408 that identifies the account identifier of the customer (e.g., the accountholder associated with customer device 406 and / or the account identifier), transaction processing system 402 may communicate the authorization request to issuer system 404 (e.g., the issuer system that issued the payment device and / or account identifier). Issuer system 404 may determine an authorization decision (e.g., approve, deny, and / or the like) based on the authorization request, and / or issuer system 404 may generate an authorization response based on the authorization decision and / or the authorization request. Issuer system 404 may communicate the authorization response to transaction processing system 402. Transaction processing system 402 may communicate the authorization response to acquirer system 41069W0191.DOCX Page 42 of 58Attorney Docket No.: 08223-2506144 (9949WO01)and / or payment gateway system 412. In some non-limiting embodiments or aspects, acquirer system 410 may communicate the authorization response to payment gateway system 412 and / or merchant system 408. Additionally or alternatively, payment gateway system 412 (and / or acquirer system 410) may communicate the authorization response to merchant system 408.

[0129] In some non-limiting embodiments or aspects, transaction processing system 402 and / or issuer system 404 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 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 be trained based on synthetic data generated, as described herein. Transaction processing system 402 and / or issuer system 404 may perform at least one task (e.g., generate a prediction and / or generate an embedding) 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, 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 402 may communicate at least one message based on performing the task (e.g., generating the prediction and / or generating an embedding) to issuer system 404 (e.g., along with the authorization request). In some non-limiting embodiments or aspects, issuer system 404 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 generation of the embedding).

[0130] 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 customer device 406 and / or the like) and / or transaction data associated with the transaction. For example, merchant system 408 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 408 may communicate the clearing message(s) to acquirer system 410 (and / or payment gateway system 412, which may communicate the clearing message(s) to acquirer system 410). Acquirer system 410 may communicate the clearing69W0191.DOCX Page 43 of 58Attorney Docket No.: 08223-2506144 (9949WO01)message(s) to transaction processing system 402. Transaction processing system 402 may communicate the clearing message(s) to issuer system 404. Issuer system 404 may generate at least one settlement message based on the clearing message(s). In some non-limiting embodiments or aspects, issuer system 404 may communicate the settlement message(s) and / or funds to transaction processing system 402 (and / or a settlement bank system associated with transaction processing system 402), and transaction processing system 402 (and / or the settlement bank system) may communicate the settlement message(s) and / or funds to acquirer system 410. Additionally or alternatively, issuer system 404 may communicate the settlement message(s) and / or funds to acquirer system 410. In some non-limiting embodiments or aspects, acquirer system 410 may communicate the settlement message(s) and / or funds to merchant system 408 (and / or an account associated with merchant system 408).

[0131] Communication network 414 may include one or more wired and / or wireless networks. For example, communication network 414 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), a 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.

[0132] The number and arrangement of systems, devices, and / or networks shown in FIG. 4 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. 4. Furthermore, two or more systems or devices shown in FIG. 4 may be implemented within a single system or device, or a single system or device shown in FIG. 4 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 environment 400 may perform one or more functions described as being performed by another set of systems or another set of devices of environment 400.69W0191.DOCX Page 44 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0133] Referring now to FIG. 5, shown is a diagram of example components of device 500, according to non-limiting embodiments or aspects. Device 500 may correspond to at least one of model analysis system 102, database 104, and / or user device 106 in FIG. 1 and / or at least one of transaction processing system 402, issuer system 404, customer device 406, merchant system 408, and / or acquirer system 410 in FIG. 4, as an example. In some non-limiting embodiments or aspects, such systems or devices in FIG. 1 or FIG. 4 may include at least one device 500 and / or at least one component of device 500. The number and arrangement of components shown in FIG.5 are provided as an example. In some non-limiting embodiments or aspects, device 500 may include additional components, fewer components, different components, or differently arranged components than those shown in FIG. 5. Additionally or alternatively, a set of components (e.g., one or more components) of device 500 may perform one or more functions described as being performed by another set of components of device 500.

[0134] As shown in FIG. 5, device 500 may include bus 502, processor 504, memory 506, storage component 508, input component 510, output component 512, and communication interface 514. Bus 502 may include a component that permits communication among the components of device 500. In some non-limiting embodiments or aspects, processor 504 may be implemented in hardware, firmware, or a combination of hardware and software. For example, processor 504 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 506 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 504.

[0135] With continued reference to FIG. 5, storage component 508 may store information and / or software related to the operation and use of device 500. For example, storage component 508 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 510 may include a component that permits device 500 to receive information, such as via user input (e.g., a touch screen69W0191.DOCX Page 45 of 58Attorney Docket No.: 08223-2506144 (9949WO01)display, a keyboard, a keypad, a mouse, a button, a switch, a microphone, etc.). Additionally or alternatively, input component 510 may include a sensor for sensing information (e.g., a global positioning system (GPS) component, an accelerometer, a gyroscope, an actuator, etc.). Output component 512 may include a component that provides output information from device 500 (e.g., a display, a speaker, one or more light-emitting diodes (LEDs), etc.). Communication interface 514 may include a transceiver-like component (e.g., a transceiver, a separate receiver and transmitter, etc.) that enables device 500 to communicate with other devices, such as via a wired connection, a wireless connection, or a combination of wired and wireless connections. Communication interface 514 may permit device 500 to receive information from another device and / or provide information to another device. For example, communication interface 514 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.

[0136] Device 500 may perform one or more processes described herein. Device 500 may perform these processes based on processor 504 executing software instructions stored by a computer-readable medium, such as memory 506 and / or storage component 508. 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 506 and / or storage component 508 from another computer-readable medium or from another device via communication interface 514. When executed, software instructions stored in memory 506 and / or storage component 508 may cause processor 504 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.69W0191.DOCX Page 46 of 58Attorney Docket No.: 08223-2506144 (9949WO01)

[0137] 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.69W0191.DOCX Page 47 of 58

Claims

Attorney Docket No.: 08223-2506144 (9949WO01)WHAT IS CLAIMED IS:

1. A system, comprising:at least one processor configured to:receive input data for a state space machine learning model; generate an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data;assign a plurality of weights to the sequence of tokens of the input sequence;provide the input sequence to a first block of the state space machine learning model;obtain an attention matrix based on the input sequence at the first block of the state space machine learning model; anddisplay an attention pattern based on the attention matrix in a user interface.

2. The system of claim 1 , wherein the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein, when providing the input sequence to the first block of the state space machine learning model, the at least one processor is configured to:provide the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following:a cross layout ordering method;a diagonal layout ordering method;a z-curve layout ordering method;a Hilbert layout ordering method;a Peano layout ordering method; ora spiral layout ordering method.

3. The system of claim 1 , wherein the first block of the state space machine learning model is a first block of a first stage of the state space machine69W0191.DOCX Page 48 of 58Attorney Docket No.: 08223-2506144 (9949WO01)learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the at least one processor is further configured to:obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model;combine the first attention matrix and the second attention matrix to provide a combined attention matrix; andapply a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix;wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to:display an attention pattern based on the reduced dimension combined attention matrix in the user interface.

4. The system of claim 1, wherein the input data comprises a first data record and a second data record, and wherein, when generating the input sequence for the state space machine learning model, the at least one processor is configured to:generate a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; and generate a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; andwherein, when assigning the plurality of weights to the sequence of tokens of the input sequence, the at least one processor is configured to:assign a first plurality of weights to the first sequence of tokens of the first input sequence; andassign a second plurality of weights to the second sequence of tokens of the second input sequence;wherein, when providing the input sequence to the first block of the state space machine learning model, the at least one processor is configured to:provide the first input sequence to the first block of the state space machine learning model; and69W0191.DOCX Page 49 of 58Attorney Docket No.: 08223-2506144 (9949WO01)provide the second input sequence to the first block of the state space machine learning model;wherein, when obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model, the at least one processor is configured to:obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model; andobtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model.

5. The system of claim 4, wherein the at least one processor is further configured to:aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix; andapply a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; and wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to:display an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

6. The system of claim 1, wherein the at least one processor is further configured to:adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.

7. The system of claim 1, wherein the at least one processor is further configured to:receive a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and69W0191.DOCX Page 50 of 58Attorney Docket No.: 08223-2506144 (9949WO01)wherein, when displaying the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to:display the attention pattern based on the attention matrix according to the display mode in the user interface.

8. A computer-implemented method, comprising:receiving, with at least one processor, input data for a state space machine learning model;generating, with at least one processor, an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data;assigning, with at least one processor, a plurality of weights to the sequence of tokens of the input sequence;providing, with at least one processor, the input sequence to a first block of the state space machine learning model;obtaining, with at least one processor, an attention matrix based on the input sequence at the first block of the state space machine learning model; and displaying, with at least one processor, an attention pattern based on the attention matrix in a user interface.

9. The computer-implemented method of claim 8, wherein the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein providing the input sequence to the first block of the state space machine learning model comprises:providing the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following:a cross layout ordering method;a diagonal layout ordering method;a z-curve layout ordering method;a Hilbert layout ordering method;a Peano layout ordering method; ora spiral layout ordering method.69W0191.DOCX Page 51 of 58Attorney Docket No.: 08223-2506144 (9949WO01)10. The computer-implemented method of claim 8, wherein the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the method further comprises:obtaining a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model;combining the first attention matrix and the second attention matrix to provide a combined attention matrix; andapplying a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises:displaying an attention pattern based on the reduced dimension combined attention matrix in the user interface.

11. The computer-implemented method of claim 8, wherein the input data comprises a first data record and a second data record, and wherein generating the input sequence for the state space machine learning model comprises:generating a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record; and generating a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; andwherein assigning the plurality of weights to the sequence of tokens of the input sequence comprises:assigning a first plurality of weights to the first sequence of tokens of the first input sequence; andassigning a second plurality of weights to the second sequence of tokens of the second input sequence;wherein providing the input sequence to the first block of the state space machine learning model comprises:69W0191.DOCX Page 52 of 58Attorney Docket No.: 08223-2506144 (9949WO01)providing the first input sequence to the first block of the state space machine learning model; andproviding the second input sequence to the first block of the state space machine learning model;wherein obtaining the attention matrix based on the input sequence at the first block of the state space machine learning model comprises:obtaining a first attention matrix based on the first input sequence at the first block of the state space machine learning model; andobtaining a second attention matrix based on the second input sequence at the first block of the state space machine learning model.

12. The computer-implemented method of claim 11, wherein the method further comprises:aggregating the first attention matrix and the second attention matrix to provide an aggregated attention matrix; andapplying a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix; and wherein displaying the attention pattern based on the attention matrix in the user interface comprises:displaying an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

13. The computer-implemented method of claim 8, wherein the method further comprises:adjusting one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.

14. The computer-implemented method of claim 8, wherein the method further comprises:receiving a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; and69W0191.DOCX Page 53 of 58Attorney Docket No.: 08223-2506144 (9949WO01)wherein displaying the attention pattern based on the attention matrix in the user interface comprises:displaying the attention pattern based on the attention matrix according to the display mode in the user interface.

15. 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 input data for a state space machine learning model; generate an input sequence for the state space machine learning model based on the input data, wherein the input sequence comprises a sequence of tokens associated with the input data;assign a plurality of weights to the sequence of tokens of the input sequence;provide the input sequence to a first block of the state space machine learning model;obtain an attention matrix based on the input sequence at the first block of the state space machine learning model; anddisplay an attention pattern based on the attention matrix in a user interface.

16. The computer program product of claim 15, wherein the sequence of tokens of the input sequence comprises a plurality of patches, wherein each patch of the plurality of patches is associated with a column and a row of a grid, wherein the grid represents the input data, and wherein, the program instructions that cause the at least one processor to provide the input sequence to the first block of the state space machine learning model, cause the at least one processor to:provide the input sequence to the first block of the state space machine learning model according to a patch ordering method, wherein the patch ordering method comprises at least one of the following:a cross layout ordering method;a diagonal layout ordering method;a z-curve layout ordering method;a Hilbert layout ordering method;69W0191.DOCX Page 54 of 58Attorney Docket No.: 08223-2506144 (9949WO01)a Peano layout ordering method; ora spiral layout ordering method.

17. The computer program product of claim 15, wherein the first block of the state space machine learning model is a first block of a first stage of the state space machine learning model, wherein the attention matrix based on the first block is a first attention matrix, and wherein the program instructions further cause the at least one processor to:obtain a second attention matrix based on the input sequence at a second block of the first stage of the state space machine learning model;combine the first attention matrix and the second attention matrix to provide a combined attention matrix; andapply a dimensionality reduction (DR) algorithm to the combined attention matrix to provide a reduced dimension combined attention matrix; and wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, cause the at least one processor to:display an attention pattern based on the reduced dimension combined attention matrix in the user interface.

18. The computer program product of claim 15, wherein the input data comprises a first data record and a second data record, and wherein, the program instructions that cause the at least one processor to generate the input sequence for the state space machine learning model, cause the at least one processor to:generate a first input sequence for the state space machine learning model based on the first data record of the input data, wherein the first input sequence comprises a first sequence of tokens associated with the first data record;generate a second input sequence for the state space machine learning model based on the second data record of the input data, wherein the second input sequence comprises a second sequence of tokens associated with the second data record; andwherein, the program instructions that cause the at least one processor to assign the plurality of weights to the sequence of tokens of the input sequence, cause the at least one processor to:69W0191.DOCX Page 55 of 58Attorney Docket No.: 08223-2506144 (9949WO01)assign a first plurality of weights to the first sequence of tokens of the first input sequence; andassign a second plurality of weights to the second sequence of tokens of the second input sequence;wherein, the program instructions that cause the at least one processor to provide the input sequence to the first block of the state space machine learning model, cause the at least one processor to:provide the first input sequence to the first block of the state space machine learning model; andprovide the second input sequence to the first block of the state space machine learning model;wherein, the program instructions that cause the at least one processor to obtain the attention matrix based on the input sequence at the first block of the state space machine learning model, cause the at least one processor to:obtain a first attention matrix based on the first input sequence at the first block of the state space machine learning model; andobtain a second attention matrix based on the second input sequence at the first block of the state space machine learning model;aggregate the first attention matrix and the second attention matrix to provide an aggregated attention matrix; andapply a dimensionality reduction (DR) algorithm to the aggregated attention matrix to provide a reduced dimension aggregated attention matrix;wherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, cause the at least one processor to:display an attention pattern based on the reduced dimension aggregated attention matrix in the user interface.

19. The computer program product of claim 15, wherein the program instructions further cause at least one processor to:adjust one or more weights of the plurality of weights assigned to the sequence of tokens of the input sequence during a training procedure, wherein each weight of the plurality of weights is associated with an indication of importance of a token of the sequence of token relative to other tokens of the sequence of tokens.69W0191.DOCX Page 56 of 58Attorney Docket No.: 08223-2506144 (9949WO01)20. The computer program product of claim 15, wherein the program instructions further cause the at least one processor to:receive a selection of a display mode for displaying the attention pattern in the user interface, wherein the display mode comprises a scatterplot view mode; andwherein, the program instructions that cause the at least one processor to display the attention pattern based on the attention matrix in the user interface, the at least one processor is configured to:display the attention pattern based on the attention matrix according to the display mode in the user interface.69W0191.DOCX Page 57 of 58

Images