Supervised VAE for optimization of value function and generation of expectation data

Abstract

A model learning and sample value generation framework includes systems and methods for synthesizing integrated encoding, decoding, and value prediction, and optimization functions to reconstruct an original input sample data space as accurately as possible. The system utilizes a variational autoencoder model to generate samples of the data space as realistic as possible. The system learns a value prediction function based on latent feature data rather than the original input data to achieve a target result. Furthermore, the system solves an optimization problem in the latent space without constraints, thereby avoiding the difficulty of optimizing in the original sample data space. The generated optimal samples are as similar as possible to real-world input samples. The system provides a flexible data generation mechanism that is applicable to a variety of target result specifications.

Description

Technical Field

[0001] This invention relates to the application of machine learning and artificial intelligence models to solve the problem of generating optimal samples for achieving target results, and to using variational autoencoder (VAE) models to generate data space samples that are as realistic as possible. Background Technology

[0002] In artificial intelligence (AI) applications, a very common need is to generate optimal data samples as input for training machine learning models to achieve specific target results. For example, healthcare providers, such as hospitals, need to optimize the utilization of their resources to achieve optimal hospital performance. In this case, the resource utilization "real-world" input sample data is the data that needs to be optimized when hospital performance is the target result. As another example, a primary school teacher needs to optimize an essay to present a model essay with a score of 100 to his / her students. In this case, the real-world essay input data is the data that needs optimization, and the score is the target result.

[0003] However, generating such desired input data samples presents several challenges. First, the relationship between the input data samples and the output target is not predefined and needs to be learned from large datasets. In other words, the first challenge lies in modeling the relationship between the input data x and the output data y.

[0004] Secondly, the space of input x is usually high-dimensional and not well defined. Therefore, direct optimization in space x is infeasible, and it is difficult to give explicit constraints on x.

[0005] Third, the generated optimal sample space corresponding to the input x should be as similar as possible to real-world samples. The generated samples should be sampled from the same distribution as the real-world data to maintain consistency with the real-world data.

[0006] Existing methods cannot simultaneously address all three challenges mentioned above. Summary of the Invention

[0007] The following description is intended to be illustrative only. It is not intended to limit the scope of the claims.

[0008] According to one aspect, this disclosure provides a system and method for generating optimal data samples for a machine learning model to achieve a specific target result when the relationship between input samples and output targets is not predefined and needs to be learned from a large dataset, i.e., when it is necessary to model the relationship between input data x and output data y.

[0009] According to another aspect, this disclosure provides a system and method for generating optimal samples for a specific objective result when the space of input data x is high-dimensional and not clearly defined, making any direct optimization in the space x infeasible and making it difficult to give explicit constraints on x.

[0010] According to another aspect, this disclosure provides a system and method for generating optimal samples for a specific target result, such that the generated optimal sample x* is as similar as possible to a real-world sample, that is, the generated sample should be sampled from the same distribution as the real-world data in order to maintain consistency with the real-world data.

[0011] According to one aspect of the present invention, a computer-implemented method is provided for generating optimal model input data for achieving a target result. The method includes: generating a latent feature representation of input data in a latent feature space using an encoder model of a supervised variational autoencoder (VAE); receiving a VAE decoder model to learn to reconstruct the input data using the latent feature representation of the input data; receiving a value predictor model to learn a relationship between the input data and a target result using the latent feature representation of the input data; simultaneously training the VAE decoder model and the value predictor model; optimizing the latent feature space representation of the input data using the trained value predictor model; receiving the optimized latent feature space representation of the input data at the trained VAE decoder model; and running the trained VAE decoder model to generate optimized samples of the input data for achieving the target result based on the optimized latent feature space representation of the input data.

[0012] According to one aspect, a computer system is provided for generating optimal model input data for achieving a target result. The computer system includes: a memory storage device for storing a computer-readable program; and at least one processor adapted to run the computer-readable program to configure the at least one processor to: generate a latent feature representation of input data in a latent feature space using an encoder model of a supervised variational autoencoder (VAE); receive a VAE decoder model to learn to reconstruct the input data using the latent feature representation of the input data; receive a value predictor model to learn a relationship between the input data and a target result using the latent feature representation of the input data; simultaneously train the VAE decoder model and the value predictor model; optimize the latent feature space representation of the input data using the trained value predictor model; receive the optimized latent feature space representation of the input data at a trained VAE decoder; and run the trained VAE decoder to generate optimal samples of the input data for achieving the target result based on the optimized latent feature space representation of the input data.

[0013] On the other hand, a computer program product for performing operations is provided. This computer program product includes a storage medium capable of being read by processing circuitry and storing instructions for executing a method by the processing circuitry. The method is the same as that listed above. Attached Figure Description

[0014] The foregoing aspects and other features are explained in the following description in conjunction with the accompanying drawings, wherein:

[0015] Figure 1 An exemplary computer system is schematically illustrated, which is adapted to implement an embodiment of the invention for automatically generating optimal samples for a specific target result;

[0016] Figure 2 The various system block diagrams depict a two-stage framework for generating optimal samples for a specific target result according to an embodiment of the present invention.

[0017] Figure 3 A method for running a two-stage framework, invoked by a computer system according to an embodiment, is shown, the two-stage framework being used to generate optimal samples for a specific target outcome;

[0018] Figure 4 An example table is depicted according to an embodiment of the present invention, which indicates the example optimal sample data input / output and optimized target result generated by two-stage network processing of an example resource utilization optimization problem;

[0019] Figure 5A schematic diagram of an example computer or processing system according to an embodiment of the present invention is shown;

[0020] Figure 6 A cloud computing environment according to an embodiment of the present invention is described; and

[0021] Figure 7 An abstract model layer according to an embodiment of the present invention is described. Detailed Implementation

[0022] According to embodiments, this disclosure provides a system and method for rapidly and automatically generating optimal samples for machine learning models to achieve target results. A supervised variational autoencoder method is implemented to solve the problem, generating optimal sample data that will yield the optimal target result (optimal value function).

[0023] In this embodiment, the system and method are implemented in two phases: a learning phase and a generation phase. In the learning phase, a supervised variational autoencoder (VAE) is trained to simultaneously learn: (1) the distribution of the input data x; and (2) the relationship between the input data x and the target output y. In the generation phase, an optimizer is used to solve an unconstrained optimization problem in the latent space z to generate various data for different purposes. The system learns a value prediction function based on latent features rather than the original input data. This system provides a flexible data generation mechanism applicable to various target outcome specifications.

[0024] like Figure 1 As shown, in the context of solving a problem according to one embodiment to generate optimal sample data that will lead to the optimal target result, the tool 100 implementing the system and method is a computer system, computing device, mobile device, or server. In some aspects, computing device 100 may include, for example, a personal computer, laptop computer, tablet computer, smart device, smartphone, or any other similar computing device.

[0025] The computing system 100 includes one or more hardware processors 152A, 152B, memory 150 for storing operating systems, application programming interfaces (APIs), and program instructions, a network interface 156, a display device 158, an input device 159, and any other features common to computing devices. In some aspects, the computing system 100 may be, for example, any computing device configured to communicate via a public or private communication network 99 with one or more websites 125, including a web-based or cloud-based server 120. In an example implementation, such data may include hospital resource utilization data, which can be used to address how to minimize hospital labor costs. Such data may be stored electronically in a database 130.

[0026] In addition, as shown as part of system 100, local memory is provided for use in the data processing framework, which may include attached memory storage device 160 or remote memory storage device, such as a database, which can be accessed via a remote network connection for input to system 100.

[0027] exist Figure 1 In the illustrated embodiment, processors 152A and 152B may include, for example, a microcontroller, a field-programmable gate array (FPGA), or any other processor configured to perform various operations. A communication channel 140, such as a wired connection like a data bus, address bus, input / output (I / O) data line, video bus, expansion bus, etc., is also shown for routing signals between the various components of system 100. Processors 152A and 152B are configured to execute instructions described below. These instructions may be stored, for example, as programming modules in a separate associated memory storage device 150.

[0028] Memory 150 may include non-transitory computer-readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory or others. Memory 150 may include, for example, other removable / non-removable, volatile / non-volatile storage media. By way of non-limiting example only, memory 150 may include a portable computer disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable optical disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing.

[0029] Network interface 156 is configured to send data or information to and receive data or information from web server 120, for example, via a wired or wireless connection. For example, network interface 156 may utilize wireless technologies and communication protocols, such as… WIFI (e.g., 802.11a / b / g / n), cellular networks (e.g., CDMA, GSM, M2M and 3G / 4G / 4G LTE, 5G), near-field communication systems, satellite communication, communication via a local area network (LAN), communication via a wide area network (WAN), or any other form of communication that allows computing device 100 to send or receive information from server 120.

[0030] Display 158 may include, for example, a computer monitor, a television, a smart television, a display screen integrated into a personal computing device, such as a laptop computer, a smartphone, a smartwatch, a virtual reality headset, a smart wearable device, or any other mechanism for displaying information to a user. In some aspects, display 158 may include a liquid crystal display (LCD), an electronic paper / electronic ink display, an organic LED (OLED) display, or other similar display technologies. In some aspects, display 158 may be touch-sensitive and may also function as an input device.

[0031] Input device 159 may include, for example, a keyboard, mouse, touch-sensitive display, keypad, microphone, or other similar input devices, or any other input devices that can be used alone or together to provide a user with the ability to interact with computing device 100. In an embodiment, through a user interface, a user can input a specific target outcome aimed at solving a problem. For example, in the context of solving a problem related to hospital resource utilization, a user can specify a target outcome y, such as how to minimize the hospital's labor costs.

[0032] Regarding the configuration of the computer system to analyze and process data to generate an optimal sample (potential) space for building a model to generate the target output, local or remote storage 160 may be configured to temporarily store data or information 162 related to the problem to be solved in a specific domain, such as hospital resource utilization data or other data obtained from a remote location (e.g., a web server).

[0033] This data is stored in a database and processed to generate an optimal sample space for addressing upcoming problems in specific fields (e.g., healthcare, education, etc.).

[0034] The captured data 162 may be data mined from information stored in electronic database 130 or other data sources (not shown). Alternatively, this data may be stored in a separate local storage device attached to computer system 100.

[0035] like Figure 1 As shown, the memory 150 of the computer system 100 also stores a processing module, which includes programming instructions executed by a processor, the programming instructions being adapted to configure the computer system to provide an optimal potential sample space for solving a problem using VAE technology.

[0036] In one embodiment, one of the programming processing modules stored in the associated memory 150 includes a data ingestion module 165, which provides instructions and logic for operating circuitry to access / receive data (e.g., structured or unstructured) and presenting them as input data x for use by other modules that process the data according to embodiments of the invention.

[0037] In this embodiment, the VAE encoder module 170 provides instructions and logic for operating the circuitry to receive input data content related to the problem to be solved. This data may be an n-dimensional vector of information about a predicted problem to be solved. The VAE encoder encodes the latent properties of the input in a probabilistic (distribution) manner, thereby generating a latent sample space. The encoder (E) captures the distribution of the input data x and encodes the input x into the decoded latent space z.

[0038] In an embodiment, the VAE encoder module 170 can run encoding functions modeled according to a deep neural network (DNN) model pipeline architecture, including but not limited to: convolutional neural network (CNN), recurrent neural network (RNN) or multilayer perceptron (MLP) model pipelines, and possible combinations and variations thereof 175, which can perform encoding of input data x depending on the type of problem to be solved.

[0039] Another programming processing module stored in the associated memory 150 of system 100 includes a VAE decoder module 180, which uses logic and instructions to operate circuitry that provides decoder functionality to reconstruct data x within an optimal error range based on the input latent sample space z. In embodiments, the decoder (D) module 180 can run decoding functions modeled according to a deep neural network model architecture, including but not limited to: CNN, RNN, or MLP model pipelines, and possible combinations and variations thereof 185, which can perform decoding of the input latent sample space data z depending on the type of problem to be solved.

[0040] Another processing module stored in the associated computer memory 150 includes a value predictor module 190, which uses logic and instructions to operate circuitry to run a function (V) for predicting a target outcome to a specified precision. In embodiments, the value predictor module 190 also runs a deep neural network model architecture, including but not limited to: CNN, RNN, or MLP model pipelines and possible combinations and variations thereof, configured to predict the value of the target outcome as accurately as possible based on the latent feature space z.

[0041] Another processing module stored in the associated computer memory 150 includes an optimizer module 195, which uses logic and instructions to operate circuitry to solve specific optimization problems. The optimizer runs an optimization algorithm (O) to solve the unconstrained optimization problem, generating an optimized latent sample space z* to obtain the best result for the specified objective outcome in the prediction problem being solved. The optimization algorithm can be any common algorithm, such as stochastic gradient descent (SGD). Several types of optimization can be performed; the only difference between them is the formula for the optimization objective.

[0042] like Figure 1 As further shown, the memory 150 includes a supervisory program 110 with instructions for configuring the computing system 100 to invoke each program module and to supervise the operation of the VAE learning and generation phase, which optimizes the latent sample space z* to obtain the best result for a specified target outcome in the prediction problem being solved, i.e., generating the best sample with the optimal target outcome. The supervisory program configures the computer system to generate the optimal target outcome value, which is represented as the data value x* obtained by decoding the optimal latent sample space z*, i.e., x* = D(z*).

[0043] Figure 2 A system for implementing a two-stage framework 200 is described, which is used to generate optimized latent variables z* in the latent variable space z as a result of processing input data x using a supervised VAE in the first stage and an optimizer (O) in the second stage.

[0044] In the first learning phase 205, input data x 210, comprising one or more vector forms of attributes related to the specific problem to be solved, is input to the encoder model (E) 175 of the supervised VAE. Based on the input attribute data vector x, the encoder DNN model (E) is trained to capture the distribution of the input data x by encoding the input x into the unwound latent variable space z, i.e., E(x) → z. The encoder takes sample data x as input and then generates a feature representation z = E(x) in the latent space 220. Depending on the type of input data x, the embodiment of the encoder function (E) can be a CNN, RNN, or MLP. The benefit of encoding is that it transforms the input in a complex high-dimensional space into the unwound low-dimensional latent space 220.

[0045] Continue, as Figure 2As shown, the decoder model 185 receives the sample latent space z 220 as input and decodes it to reconstruct the sample x, i.e., D(z)→x, within a predetermined loss or error metric determined by the loss function. Depending on the input data type x, the implementation of the decoder function (D) can be a CNN, RNN, or MLP. As a result of the decoder model processing, the method generates an unsupervised reconstruction error loss |D(E(x))-x|.

[0046] Simultaneously, the value predictor model (V) 190 receives the same latent space z 220 as input and predicts the value of the target outcome y 230 as accurately as possible, or predicts the value of the target outcome y 230 within a predetermined loss or error metric. Therefore, in addition to learning the distribution of the input data x using the latent variable space z 220, this method also learns the relationship between the input data x and the target outcome output y, i.e., x→z→(x,y), within a predetermined loss or error metric determined by the loss function. Depending on the type of input data x, the embodiment of the value predictor function (V) can be a CNN, RNN, or MLP designed to model the relationship between the target outcome and the input data. However, it is not a direct function from x to y. It is a two-step function from x to z and then to y. This part generates the supervised prediction error loss |V(E(x))-y| for training the VAE. Given these operations, the method can perform downstream optimal sample generation for the target outcome.

[0047] During the training of the VAE model in a manner that predicts the outcome of a given sample latent space, the loss to be minimized is alternatively expressed according to the following equation 1):

[0048]

[0049] Where ||D(E(x) i ))-x i || represents the loss component that represents the reconstruction loss of the input data x, ||y i -V(E(x i ))|| represents the loss component of the label prediction loss.

[0050] More specifically, in the first learning phase 205, the loss function of the supervised VAE consists of three parts: 1) reconstruction loss, i.e., E(log P(x|z)). This part can be further divided into two parts for continuous features and discrete features; 2) prior loss, i.e., D_KL(Q(z|x)|P(z)). This part calculates the KL divergence distance between the posterior z|x and the prior p(z); and 3) prediction loss, i.e., E(log V(y|z)). In one embodiment, this part is a supervised loss that calculates the squared difference between y|z and the true y. These loss functions can be customized according to specific applications. That is, this method is an effective solution for such a wide range of similar tasks, and different tasks can have different loss functions.

[0051] like Figure 2 As further illustrated, a system for implementing a two-stage framework 200 is depicted, which is used to generate optimized latent variables z* in the latent variable space z as a result of processing input data x using a supervised VAE in the first and second learning stages 250, to run optimization techniques for solving unconstrained optimization problems in the latent variable space z, and to provide various data generation methods for different purposes.

[0052] In this embodiment, after the learning phase 205 is completed, in the generation phase 250, the system runs the optimization module 195 to run an optimizer for finding the optimal latent space variable z*260 for a specific target result.

[0053] Different types of optimizers are used depending on the specific type of optimization problem being solved. The only difference between optimizers lies in the formula of the optimization objective. The optimization algorithm can be any common algorithm, such as stochastic gradient descent (SGD), Adam, or AdaGrad.

[0054] In this embodiment, the first type of optimizer (O1) solves the global optimization as defined by the following formula:

[0055] z*=arg max z V(z).

[0056] Where V(z) is the value predictor model. Using the optimizer (O1), the task is to perform unconstrained optimization to find the best sample that generates the maximum objective result value.

[0057] In another embodiment, the second optimizer (O2) solves for local optimization given specific input data x as defined by the following formula:

[0058] z x * = arg max z V(z)-γ||D(z)-x||

[0059] Where V(z) is the value predictor function, D(z) is the decoder function applied to the corresponding latent space variable z, x is the specific input data value, and γ is a coefficient used to control the influence of the regularization term ||D(z)-x||. For example, the coefficient γ used to control the influence of the regularization term can be 0.1 or 0.5 or other values. Using an optimizer (O2), the unconstrained optimization problem is assigned the task of generating optimal samples that are more like the given input x but have a larger objective outcome.

[0060] In another embodiment, a third-type optimizer (O3) solves local optimizations given a target outcome value y, which is defined by the following formula:

[0061] z y *=arg min z ||V(z)-y||

[0062] Where V(z) is the value predictor function, and y is the predicted result value. Using the optimizer (O3), the task is an unconstrained optimization problem to generate the optimal sample that is consistent with the objective result value y.

[0063] As a result of solving the optimization problem using any optimizer (O1)-(O3), for each potential space variable z, a corresponding optimized potential variable space z* is generated.

[0064] Continue, in Figure 2 In this process, the optimized latent space variable z* is input to the supervised VAE decoder 180 to generate an optimized input data variable x*, i.e., D(z*)→x*, where x* is the final generated output sample. For example, given an input vector including variables x for the prediction problem to be solved, after running the first and second stages, a corresponding vector of optimized input variables x* is generated, i.e., is computed to produce the optimized target result y.

[0065] That is, by inputting the optimized latent space input variable z* into the value predictor function V(z*), the predictor function will generate the optimal (best) target output value y.

[0066] Figure 3 It shows in Figure 2The method implemented in the two-stage framework 200, at the first step 302, involves the data ingestion module receiving input data related to the problem to be solved, and performing any data formatting to present it in a form used by the VAE encoder module 170. An example, non-limiting problem to be solved is in the healthcare field, specifically involving optimizing resource utilization data x in a manner that minimizes per-patient labor costs, user-specified target outcome data y. The input data may include data from multiple hospitals (e.g., 100 hospitals). Each hospital provides input pairs of input data x and y.

[0067] For example, such as Figure 4 As shown, given a healthcare provider such as a hospital, the hospital utilizes its resources in various ways, such as the number of beds, nurses, or other resources. Different configurations of these resources will result in different hospital performance. Therefore, the problem to be solved is how to optimize the utilization of its resources so that the hospital can achieve optimal performance.

[0068] In an embodiment, the system 100 of the present invention is thus assigned tasks to generate an optimal sample space with the best utilization of hospital resources, which minimizes the hospital’s labor costs.

[0069] Figure 4 Table 400 depicts data input from each hospital, including resource utilization attribute data 402 related to hospital costs (i.e., x is real-world data). In this example, data received as system input relates to hospital resource usage across multiple hospitals (e.g., 100 hospitals). In this embodiment, the real-world input data 402 for each hospital is a 20-dimensional data vector, with each dimension corresponding to a hospital resource utilization attribute, depicted as follows: Figure 4 The corresponding row 401 is shown in the table. Such data 402 from each hospital may include character-type data 405, such as binary data with values ​​"1" or "0," indicating, for example, whether the hospital has 80% or more full-time staff, whether patients have case managers, whether the hospital is a teaching hospital, or whether the head nurse provides 50% or more of the time for clinical patient care, etc. Such data 402 from each hospital may include numeric-type data 408, such as the percentage of overtime hours to working hours, average wage index, hospital case mix index value, bed capacity, equivalent average length of stay value, values ​​related to hours paid per equivalent patient day or labor costs per equivalent patient day, etc. One of the hospital resource utilization attributes is shown at row 410, which corresponds to the target outcome value 420 that is intended to be optimized (minimized), namely, labor costs per equivalent patient day.

[0070] In this embodiment, for each input data attribute value 402, a corresponding statistical value, such as the median 411 and the mean 413, is calculated based on the input data from each hospital. For example, from each of the real-world data vectors input from 100 hospitals, the median "bed capacity" attribute 415 value can be determined to be 24. The data from each hospital can be input as an m-dimensional vector, for example, as... Figure 4 The example of hospital resource utilization shows a 20-dimensional vector. During the learning phase, the system learns the distribution of the input data x.

[0071] Back Figure 3 At step 302, the user further specifies the target result y of the problem to be solved. During the learning phase, the system learns the relationship between the input data x and y. Figure 4 In the example, the objective of the example problem is to optimize the attribute at line 410, for example, minimizing the hospital's labor costs. Therefore, Figure 3 The example hospital resource utilization method is to solve the problem of determining the optimal sample x* that minimizes hospital labor costs (i.e., y* generated by the value predictor function V(z*)).

[0072] Continue to Figure 3 At 304, an operation is performed to run the VAE encoder module 170 function (E) on the received input data (e.g., a 20-dimensional data vector from each hospital), and at 307, a latent sample space variable, the latent sample space z, is generated. Depending on the task and the type of input data, the encoder model function E() (and similarly, the D() and predictor V() functions) can be any of an MLP, CNN, or RNN. For example, an MLP model-type encoder / decoder is used to process the structure type of input data such as the 20-dimensional data vector in the hospital resource utilization example.

[0073] Alternatively, the image type of the input data will be used with a CNN model-type encoder / decoder, where the input data x is the pixel data of the photo, and the target result y can be the photo's score (e.g., beauty score).

[0074] Alternatively, temporal or sequential input data will use an RNN model-type encoder / decoder. For example, in an education use case, a primary school teacher needs to optimize articles to present a model article with a score of 100 to his / her students. In such a use case, the article is the one that needs optimization, and the article score is the target result (100-point article). Given a large amount of article data to capture, and the relationship between the articles and scores, an optimal article can be generated to give a score of 100. The input x* will be the article that produces 100 article scores. This will differ from structured data because the input data x is a sequence of words and constitutes sequential data.

[0075] In this embodiment, such an encoder reduces the dimensionality of the input sample x, for example, using an MLP model-type encoder function where the encoder processes the output according to [20×5], i.e., reducing the 20-dimensional data vector in the hospital example to a 5-dimensional vector output. Supervised training is used to learn the optimal encoder / decoder model after iterative learning guided by minimizing the loss function and adjusting the initial parameters, to better reconstruct the input x data and better predict the target outcome value y.

[0076] Therefore, at 310, in parallel operation, the latent sample space attribute z is input to the VAE decoder module function (D) running at 316 to obtain the output reconstructed input sample x. At 310, the latent sample space attribute z is simultaneously input to the value predictor model (V) running at 316 to obtain the output predictor value y. The decoder model can implement a [5×20] function to ensure that the output dimension corresponds to the original input data x. The optimization (e.g., minimization) of the loss function tunes the parameters in the encoder / decoder model. In the given example, the value predictor model (V) can implement a [5×1] function to ensure that the output dimension corresponds to the desired target result y.

[0077] exist Figure 3 At position 316, the corresponding error value is generated using the loss functions of the decoder and predictor models, and then evaluated at position 318. In this embodiment, the loss function of the supervised VAE is E(log P(x|z)), where E() is the expected value of the logarithm of the conditional probability P(x|z), i.e., given the latent variable z, x i The conditional probability is given because the autoencoder reconstructs x as being more likely to be the input x. The P() function is defined using any particular probability distribution function (e.g., a Gaussian distribution). If a Gaussian distribution is used, the loss formula is transformed according to Equation 1) into a reconstruction loss function ||D(E(x)|| to be minimized for reconstructing the input data x for each data attribute i. i ))-x i In this embodiment, the reconstruction loss can be divided into two parts: one for continuous features and one for discrete features.

[0078] Another prior loss is calculated using D_KL(Q(z|x)|P(z)) to minimize the dissimilarity between the two probability distributions (Q and P). Specifically, the KL divergence loss function is calculated to measure the distance between the posterior z|x and the prior p(z), where Q(z|x) is the conditional distribution of the latent variable z given the input data x, and it corresponds to the decoder network; P(z) is the prior distribution of the latent variable z, which is assumed to follow a simple distribution, such as a standard Gaussian distribution. This prior loss term is minimized to avoid overfitting the autoencoder to the input data x.

[0079] In this embodiment, the label prediction loss for each data attribute i is calculated according to E(log V(y|z)), where V() is a probability distribution function and V(y|z) represents the probability of y given z. This loss term is the supervised loss calculated by the squared difference between y|z and the true y. The expectation is to maximize log V(y|z), i.e., E(logV(y|z)). The prediction loss term ||yV(z)|| or ||yV(E(x)|| is obtained by using a specific distribution in V().

[0080] If in Figure 3 At point 318, it is determined that the reconstruction loss and the predicted label loss are not minimized, so proceed to... Figure 3 Repeat this process for 320 to adjust the E(), D(), or V() model parameters, and return to Figure 3 Step 304 is executed again to run the VAE encoder model, decoder model, and predictor model to obtain the corresponding latent space variable z, reconstructed input sample x, and target predictor value y. Steps 304 to 318 are repeated until the error loss is minimized without overfitting.

[0081] Once the error loss is minimized to the specified accuracy at point 318, the process proceeds to point 322 to use optimizer module 190 to optimize the latent sample space samples z* generated for the target result produced by the autoencoder using optimization algorithms such as SGD, Adam, AdaGrad, etc., so that samples z* naturally resemble real-world samples. Specifically, at point 322, the optimization objective to find its maximum or minimum value is formulated. This optimization objective is problem-specific. For example, for Figure 4 The hospital resource utilization example shown has an optimization objective denoted as the labor cost attribute 420, expressed as the predictor function V(z). The optimization problem, based on the first optimizer type, is defined as min V(z), i.e., minimizing the hospital labor cost attribute. In this example, the optimal latent space variable z is expressed as z* = arg min V(z). Depending on the specific problem, the optimization objective for the expected outcome can be formulated using other types of optimizers: z* = arg max z V(z) or z x * = arg max z V(z)-γ||D(z)-x||, or, given a specific target result y, z y *=arg min z ||V(z)-y||.

[0082] exist Figure 4In the hospital resource utilization example, the expected target outcome y is the hospital labor cost of 420, which is the variable to be minimized according to z. That is, the optimization variable is the latent variable space. Therefore, the first part of the optimization objective is about the variable y, and it is solved using one of the three types of objective functions, for example, z*=arg min V(z).

[0083] In the embodiments, for Figure 3 The optimal potential sample generation stage at position 322 is optimized according to the following formula based on equation 2):

[0084] min z V(z)–logP(z)(2)

[0085] The first part 1) is the expected target result of achieving one of the three optimizer types; the second part 2) is the probability regularization term, i.e., the probability of the latent feature z*. This second regularization term ensures a high probability of the result z* and ensures the existence of decoded samples; otherwise, an extreme z* would lead to unrealistic samples x*. To regularize the variable z, it is ensured that the probability of the result z* is not too small.

[0086] Finally, Figure 3 At position 325, the method uses z* to generate the final optimal sample x* to achieve the target result y*. That is, the decoder module 180 runs the VAE decoder model using the optimized latent sample space z* to generate the final optimal sample x* to achieve the target result. Specifically, in this step, to obtain x*, the decoder function D() obtained from the learning phase operates as follows:

[0087] x*=D(z*)=D(argmin V(z))

[0088] The target result y* is obtained by calculating the following equation using the value predictor function V() obtained from the learning phase:

[0089] y*=V(E(x*))=V(z*))

[0090] return Figure 4 For hospital resource utilization example 400, the optimized objective result y*420 is the minimization of the hospital's labor costs, with a value of 95.33. That is, after obtaining the optimal resource utilization data x*, this vector is used to guide changes in 100 hospitals to reduce labor costs and lower overall costs. The vector of values ​​in the denormalized data column 425 is the optimal sample x* that will achieve this objective result.

[0091] like Figure 4As shown in Example Table 400, as a result of running the method described in this paper to construct the optimal latent sample space z*, the value predictor model will use normalized optimal result data x*423 and the corresponding denormalized optimal result data x*425 to obtain the optimized objective result output y*. The normalized result data x*423 represents the deviation of x* from the mean of the input data, e.g., greater than the mean (positive value) or less than the mean (negative value). The denormalized optimal result data x*425 represents the optimal utilization value for each attribute (row) to achieve the objective result of minimizing the hospital's labor costs. Therefore, for a specific hospital, column 425 shows the optimal configuration of resource utilization that will result in minimizing the hospital's labor costs based on the original input data vector x, which is given for all 100 hospitals based on the objective result of minimizing the hospital's labor costs or expenses. Figure 4 In Example 400 shown, for all 100 hospitals, the average value of the target attribute, labor cost per equivalent patient-day (x*), is $564.9. The optimal value for the hospital that minimizes x* is $95. The best resource utilization at the hospital that achieves this optimal denormalized value x* will be the configuration of hospital resources as shown in column 425. For example, using the example hospital resource utilization example, the x* vector provides guidance for increasing the bed capacity attribute 415 from the average of 24 to a new value of 32 to achieve the desired result of reduced labor costs.

[0092] The learning and generation framework 100 according to the embodiments herein is a system that comprehensively integrates four modules including encoding, decoding, prediction, and optimization, and utilizes a supervised VAE model to generate samples that are as realistic as possible. System 100 learns the value prediction function based on latent features rather than the original input data, which is more robust. Furthermore, the system solves the optimization problem in the latent space without constraints, avoiding the difficulties of optimization in the original space x. System 100 provides a flexible data generation mechanism suitable for various target outcome specifications.

[0093] Figure 5 An example computing system according to the present invention is illustrated. It should be understood that the described computer system is merely one example of a suitable processing system and is not intended to impose any limitation on the scope or functionality of the embodiments of the invention. For example, the system shown can be operated in many other general-purpose or special-purpose computing system environments or configurations. Suitable for Figure 5Examples of known computing systems, environments, and / or configurations used in conjunction with the systems shown may include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe systems, and distributed cloud computing environments that include any of the aforementioned systems or devices.

[0094] In some embodiments, the computer system may be described in the general context of computer system executable instructions, which are embodied as program modules stored in memory 16 and executed by the computer system. Typically, program modules may include routines, programs, objects, components, logic, data structures, etc., that perform specific tasks and / or implement specific input data and / or data types according to the present invention (see, for example, see...). Figure 3 ).

[0095] The components of the computer system may include, but are not limited to, one or more processors or processing units 12, memory 16, and a bus 14 operatively coupling various system components including memory 16 to processor 12. In some embodiments, processor 12 may execute one or more modules 11 loaded from memory 16, wherein the program modules embody software (program instructions) that cause the processor to perform one or more method embodiments of the present invention. In some embodiments, module 11 may be programmed into an integrated circuit of processor 12, loaded from memory 16, storage device 18, network 24, and / or combinations thereof.

[0096] Bus 14 can represent one or more of any of several types of bus architectures, including memory buses or memory controllers, peripheral buses, accelerated graphics ports, and processor or local buses using any of a variety of bus architectures. By way of example and not limitation, such architectures include Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MCA) buses, Enhanced ISA (EISA) buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0097] Computer systems can include a variety of computer system-readable media. Such media can be any available media accessible to the computer system, and can include volatile and non-volatile media, removable and non-removable media.

[0098] Memory 16 (sometimes referred to as system memory) may include computer-readable media in the form of volatile memory, such as random access memory (RAM), cache memory, and / or other forms. The computer system may also include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 18 may be provided for reading from and writing to non-removable, non-volatile magnetic media (e.g., a "hard disk drive"). Although not shown, a disk drive may be provided for reading from and writing to a removable, non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive may be provided for reading from or writing to a removable, non-volatile optical disk such as a CD-ROM, DVD-ROM, or other optical media. In such a case, each may be connected to bus 14 via one or more data media interfaces.

[0099] The computer system can also communicate with one or more external devices 26, such as a keyboard, pointing device, display 28, etc.; one or more devices that enable a user to interact with the computer system; and / or any device that enables the computer system to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication can occur via input / output (I / O) interface 20.

[0100] Furthermore, the computer system can communicate with one or more networks 24, such as a local area network (LAN), a general wide area network (WAN), and / or a public network (e.g., the Internet), via network adapter 22. As shown, network adapter 22 communicates with other components of the computer system via bus 14. It should be understood that, although not shown, other hardware and / or software components may also be used in conjunction with the computer system. Examples include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archive storage systems.

[0101] This invention can be a system, method, and / or computer program product at any possible level of technical detail integration. The computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to perform aspects of the invention.

[0102] Computer-readable storage media can be tangible devices capable of retaining and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices such as punch cards or recessed structures with instructions recorded thereon, and any suitable combination of the foregoing. As used herein, computer-readable storage media should not be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0103] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a suitable computing / processing device, or via a network, such as the Internet, a local area network (LAN), a wide area network (WAN), and / or a wireless network, to an external computer or external storage device. The network may include copper cables, optical fibers, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to a computer-readable storage medium within the respective computing / processing device.

[0104] Computer-readable program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages ​​(including object-oriented programming languages ​​such as Smalltalk, C++, etc.) and procedural programming languages ​​(such as the "C" programming language or similar programming languages). The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, to perform aspects of this invention, electronic circuits, including, for example, programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), may execute computer-readable program instructions to personalize the electronic circuits by utilizing the status information of the computer-readable program instructions.

[0105] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0106] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / actions specified in one or more blocks of a flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and / or other devices to operate in a particular manner, such that the computer-readable storage medium in which the instructions are stored includes an article of writing comprising instructions for implementing aspects of the functions / actions specified in one or more blocks of a flowchart and / or block diagram.

[0107] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions, which execute on the computer, other programmable apparatus or other device, perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0108] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions comprising one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions indicated in the blocks may occur in a non-consecutive order as shown in the figures. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or these blocks may sometimes be executed in reverse order, depending on the functions involved. It will also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

[0109] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the terms “comprising” and / or “including”, when used in this specification, designate the presence of stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The corresponding structures, materials, actions, and equivalents of all elements in the following claims are intended to include any structure, material, or action for performing functions in combination with other claimed elements of a particular claim. The invention has been described for purposes of illustration and description, but this description is not exhaustive or intended to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the invention. The embodiments were chosen and described in order to best explain the principles and practical application of the invention and to enable others skilled in the art to understand various embodiments of the invention with various modifications suitable for the particular purpose contemplated.

[0110] It should be understood that although this disclosure includes a detailed description of cloud computing, the implementation of the teachings set forth herein is not limited to a cloud computing environment. Rather, embodiments of the invention can be implemented in conjunction with any other type of computing environment now known or developed hereafter.

[0111] Cloud computing is a service delivery model for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services) that can be rapidly provisioned and released with minimal management effort or interaction with service providers. This cloud model may include at least five features, at least three service models, and at least four deployment models.

[0112] The features are as follows:

[0113] On-demand self-service: Cloud consumers can unilaterally and automatically provide computing power, such as server time and network storage, as needed, without requiring manual interaction with the service provider.

[0114] Wide Area Network (WAN) Access: Capabilities are available on the network and accessed through standard mechanisms that facilitate the use of heterogeneous thin or thick client platforms (e.g., mobile phones, laptops, and PDAs).

[0115] Resource pooling: A provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, where different physical and virtual resources are dynamically allocated and reallocated based on demand. Location independence has significance because consumers typically do not control or know the exact location of the resources provided, but can specify the location at a higher level of abstraction (e.g., country, state, or data center).

[0116] Rapid Flexibility: In some cases, the ability to scale outwards and inwards quickly and flexibly can be provided. For consumers, the available capacity often appears unlimited and can be purchased in any quantity at any time.

[0117] Measurement services: Cloud systems automatically control and optimize resource usage by leveraging metering capabilities at a level of abstraction appropriate to the service type (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency to both the providers and consumers of the services being utilized.

[0118] The service model is as follows:

[0119] Software as a Service (SaaS): The capability offered to consumers is the ability to use the provider's applications running on cloud infrastructure. Applications can be accessed from various client devices through thin client interfaces such as web browsers (e.g., web-based email). Consumers do not manage or control the underlying cloud infrastructure, including the network, servers, operating system, storage, or even individual application capabilities, with possible exceptions such as limited user-specific application configuration settings.

[0120] Platform as a Service (PaaS): This provides consumers with the ability to deploy consumer-created or acquired applications onto cloud infrastructure using programming languages ​​and tools supported by the provider. Consumers do not manage or control the underlying cloud infrastructure, including networks, servers, operating systems, or storage, but they have control over the deployed applications and the configuration of any application hosting environments.

[0121] Infrastructure as a Service (IaaS): This provides consumers with the capability to deliver processing, storage, networking, and other basic computing resources that enable them to deploy and run arbitrary software, which may include operating systems and applications. Consumers do not manage or control the underlying cloud infrastructure, but they do have control over the operating system, storage, deployed applications, and possibly limited control over selected networking components (e.g., host firewalls).

[0122] The deployment model is as follows:

[0123] Private cloud: Cloud infrastructure operated solely by an organization. It can be managed by the organization or a third party and can exist on-site or off-site.

[0124] Community cloud: Cloud infrastructure shared by several organizations and supporting a specific community with shared concerns (e.g., tasks, security requirements, policies, and compliance considerations). It can be managed by an organization or a third party and can exist on-site or off-site.

[0125] Public cloud: Cloud infrastructure available to the general public or large industrial groups and owned by organizations that sell cloud services.

[0126] Hybrid cloud: A cloud infrastructure is a combination of two or more clouds (private, community, or public) that remain a single entity but are bound together by standardized or proprietary technologies that enable data and applications to be ported together (e.g., cloud bursting for load balancing between clouds).

[0127] Cloud computing environments are service-oriented, focusing on statelessness, loose coupling, modularity, and semantic interoperability. At the heart of cloud computing is the infrastructure of a network of interconnected nodes.

[0128] Now for reference Figure 6The illustration depicts a cloud computing environment 50. As shown, the cloud computing environment 50 includes one or more cloud computing nodes 11 to which local computing devices used by cloud consumers can communicate, such as personal digital assistants (PDAs) or cellular phones 54A, desktop computers 54B, laptop computers 54C, and / or automotive computer systems 54N. Nodes 10 can communicate with each other. They can be physically or virtually grouped (not shown) in one or more networks, such as private clouds, community clouds, public clouds, or hybrid clouds, or combinations thereof, as described above. This allows the cloud computing environment 50 to provide infrastructure, platform, and / or software as a service, without requiring cloud consumers to maintain resources on their local computing devices. It should be understood that... Figure 6 The types of computing devices 54A-N shown are for illustrative purposes only, and computing node 10 and cloud computing environment 50 can communicate with any type of computerized device on any type of network and / or network-addressable connection (e.g., using a web browser).

[0129] Now for reference Figure 7 This demonstrates a cloud computing environment of 50 ( Figure 6 This provides a set of functional abstractions. It should be understood beforehand that... Figure 7 The components, layers, and functions shown are for illustrative purposes only, and embodiments of the invention are not limited thereto. As depicted, the following layers and corresponding functions are provided:

[0130] The hardware and software layer 60 includes hardware and software components. Examples of hardware components include: a host 61; a server 62 based on a RISC (Reduced Instruction Set Computer) architecture; a server 63; a blade server 64; a storage device 65; and network and networking components 66. In some embodiments, software components include network application server software 67 and database software 68.

[0131] The virtualization layer 70 provides an abstraction layer from which the following examples of virtual entities can be provided: virtual server 71; virtual storage 72; virtual network 73, including virtual private network; virtual application and operating system 74; and virtual client 75.

[0132] In one example, management layer 80 can provide the following functionalities: Resource Provisioning 81 provides dynamic procurement of computing resources and other resources used to perform tasks within the cloud computing environment. Metering and Pricing 82 provides cost tracking for utilizing resources within the cloud computing environment, as well as billing or invoicing for consuming these resources. In one example, these resources may include application software licenses. Security provides authentication for cloud consumers and tasks, and protection for data and other resources. User Portal 83 provides access to the cloud computing environment for consumers and system administrators. Service Level Management 84 provides cloud resource allocation and management to ensure the required service level is met. Service Level Agreement (SLA) Planning and Fulfillment 85 provides pre-scheduling and procurement of cloud resources, where future needs are anticipated according to the SLA.

[0133] Workload layer 90 provides examples of functionalities that can be leveraged in a cloud computing environment. Examples of workloads and functionalities that can be provided from this layer include: mapping and navigation 91; software development and lifecycle management 92; virtual classroom education delivery 93; data analysis and processing 94; transaction processing 95; and processing 96 for automatically generating optimal samples for target results according to aspects of this disclosure.

Claims

1. A computer-implemented method for generating optimal model input data for achieving a target result, the method comprising: The input data related to the prediction problem to be solved is received at the VAE encoder model of the supervised variational autoencoder (VAE). The VAE encoder model is used to generate a latent feature space representation of the input data in the latent feature space; The latent feature space representation of the input data is received at the VAE decoder model of the VAE; The VAE decoder model is used to learn and reconstruct the input data using the latent feature space representation of the input data; The latent feature space representation of the input data is received at the value predictor model, which is trained to learn the relationship between the input data and the target result using the latent feature space representation of the input data. Simultaneously train the VAE decoder model and the value predictor model; This forms an unconstrained optimization problem in the latent feature space. The unconstrained optimization problem is solved using a trained value predictor model to generate an optimized latent feature space representation of the input data that achieves the target outcome specified in the prediction problem. The optimized latent feature space representation of the input data is received at the trained VAE decoder model, and The trained VAE decoder model is run to generate optimal samples of the input data for achieving the target result, based on the optimized latent feature space representation of the input data.

2. The computer-implemented method of claim 1, wherein the VAE encoder model, the VAE decoder model, and the value predictor model comprise a machine learning deep neural network model selected from: a convolutional neural network (CNN), a recurrent neural network (RNN), or a multilayer perceptron (MLP).

3. The computer-implemented method according to claim 1, wherein the simultaneous training of the VAE decoder model and the value predictor model optimizes the loss function, the loss function including a reconstruction error loss component for training the VAE decoder and a label prediction error loss component for training the value predictor model.

4. The computer-implemented method of claim 1, wherein the unconstrained optimization problem is a global optimization to find an optimized latent feature space representation of the input data that generates the maximum target result value.

5. The computer-implemented method of claim 1, wherein the unconstrained optimization problem is a local optimization to find an optimized latent feature space representation of the input data that is consistent with the target result value.

6. The computer-implemented method according to claim 1, wherein the unconstrained optimization problem is a local optimization given specific input data to find an optimal sample similar to the given input data but with a larger objective result.

7. The computer-implemented method of claim 1, wherein the unconstrained optimization problem includes a probabilistic regularization component to optimize the probabilities of the latent feature space representation.

8. A computer system for generating optimal model input data for achieving a target result, the computer system comprising: Memory storage device for storing computer-readable programs, and At least one processor adapted to run the computer-readable program to configure the at least one processor to: The input data related to the prediction problem to be solved is received at the VAE encoder model of the supervised variational autoencoder (VAE). The VAE encoder model is used to generate a latent feature space representation of the input data in the latent feature space. The latent feature space representation of the input data is received at the VAE decoder model of the VAE; The VAE decoder model is used to learn and reconstruct the input data using the latent feature space representation of the input data; The latent feature space representation of the input data is received at the value predictor model, which is trained to learn the relationship between the input data and the target result using the latent feature space representation of the input data. Simultaneously train the VAE decoder model and the value predictor model; This forms an unconstrained optimization problem in the latent feature space. The unconstrained optimization problem is solved using a trained value predictor model to generate an optimized latent feature space representation of the input data that achieves the target outcome specified in the prediction problem. The optimized latent feature space representation of the input data is received at the trained VAE decoder model, and The trained VAE decoder model is run to generate optimal samples of the input data for achieving the target result, based on the optimized latent feature space representation of the input data.

9. The computer system of claim 8, wherein the VAE encoder model, the VAE decoder model, and the value predictor model comprise a machine learning deep neural network model selected from: a convolutional neural network (CNN), a recurrent neural network (RNN), or a multilayer perceptron (MLP).

10. The computer system of claim 8, wherein, in order to simultaneously train the VAE decoder model and the value predictor model, the at least one processor is further configured to optimize a loss function, the loss function including a reconstruction error loss component for training the VAE decoder and a label prediction error loss component for training the value predictor model.

11. The computer system of claim 8, wherein the unconstrained optimization problem is a global optimization to find an optimized latent feature space representation of the input data that generates the maximum target result value.

12. The computer system of claim 8, wherein the unconstrained optimization problem is selected from one of the following: local optimization to find an optimized latent feature space representation of the input data that is consistent with the target result value, or local optimization given specific input data to find an optimal sample that is similar to the given input data but has a larger target result.

13. The computer system of claim 8, wherein the unconstrained optimization problem includes: A probabilistic regularization component is used to optimize the probabilities of the latent feature space representation.

14. A computer program product comprising a computer-readable storage medium therein storing a computer-readable program, wherein the computer-readable program, when executed on a computer including at least one processor, causes the at least one processor to: The input data related to the prediction problem to be solved is received at the VAE encoder model of the supervised variational autoencoder (VAE). The VAE encoder model is used to generate a latent feature space representation of the input data in the latent feature space. The latent feature space representation of the input data is received at the VAE decoder model of the VAE; The VAE decoder model is used to learn and reconstruct the input data using the latent feature space representation of the input data; The latent feature space representation of the input data is received at the value predictor model, which is trained to learn the relationship between the input data and the target result using the latent feature space representation of the input data. Simultaneously train the VAE decoder model and the value predictor model; Form an unconstrained optimization problem in the latent feature space; solve the unconstrained optimization problem using a trained value predictor model to generate an optimized latent feature space representation of the input data that achieves the target result specified in the prediction problem; The optimized latent feature space representation of the input data is received at the trained VAE decoder model, and The trained VAE decoder model is run to generate optimal samples of the input data for achieving the target result, based on the optimized latent feature space representation of the input data.

15. The computer program product of claim 14, wherein, in order to simultaneously train the VAE decoder model and the value predictor model, the computer-readable medium further configures the at least one processor to optimize a loss function, the loss function including a reconstruction error loss component for training the VAE decoder and a label prediction error loss component for training the value predictor model.

16. The computer program product of claim 14, wherein the unconstrained optimization problem is a global optimization to find an optimized latent feature space representation of the input data that generates the maximum target result value.

17. The computer program product of claim 14, wherein the unconstrained optimization problem is selected from one of the following: local optimization to find an optimized latent feature space representation of the input data that is consistent with the target result value, or local optimization given specific input data to find an optimal sample similar to the given input data but with a larger target result.

18. The computer program product of claim 14, wherein the unconstrained optimization problem includes: A probabilistic regularization component is used to optimize the probabilities of the latent feature space representation.

Images