Systems, methods, and computer program products for data enhancement processes

Abstract

The present invention is entitled "Systems, methods and computer program products for data enhancement processes". The present invention provides technology for autonomous data enhancement. For example, one or more embodiments described herein may relate to a system including a memory that can store computer executable components. The system may also include a processor that can be operably coupled to the memory, and the processor executes the computer executable components stored in the memory. The computer executable components may include a data enhancement component that executes a random one-dimensional enhancement algorithm to enhance a data set used to train a machine learning model via multiple enhancement operations. The random one-dimensional enhancement algorithm may adopt a global enhancement parameter that defines: the distortion amplitude associated with the multiple enhancement operations, and the number of enhancement operations included in the multiple enhancement operations.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit and priority of U.S. Provisional Application No. 63 / 154,033, entitled “AUTOMATED DATA AUGMENTATION INDEEP LEARNING,” filed on February 26, 2021. The entire contents of the above application are hereby incorporated by reference herein. Technical Field

[0003] The present disclosure relates to one or more computer-implemented methods and / or systems for autonomous data augmentation to facilitate deep learning models, and more particularly, to random one-dimensional data augmentation that can improve the performance of machine learning models with minimal computational resources. Background Art

[0004] Data augmentation is used to improve deep learning model performance metrics without incurring additional computational cost at inference time. Unfortunately, creating data augmentation strategies often requires human expertise and / or domain knowledge, which is inconvenient during initial development and when transferring existing strategies between different tasks. To overcome these shortcomings, attempts have been made to automate the data augmentation process.

[0005] Typical automated data augmentation procedures employ augmentation parameters that are jointly optimized with the neural network parameters during training, which can introduce a large search space and, in turn, can significantly increase the time required to train a model. For example, one or more automated augmentation strategies have been used for models of size 10 32 The search space of employs reinforcement learning ("RL"), which may take thousands of graphics processor unit ("GPU") hours to find a solution to a single task. In addition, due to the complexity of implementing a joint optimization algorithm, an automatic reinforcement strategy may be undesirable.

[0006] An example of a typical boosting strategy is random boosting, which is an automated boosting strategy developed to solve problems with large search spaces. Random boosting eliminates the need for strategy optimization used by other techniques. For example, random boosting can reduce the search space from 10 32 Down to 10 2 However, random augmentation may require up to 100 full model training iterations to select an ideal configuration. The computational cost of performing so many training tasks can often be prohibitive. To overcome the computational cost, human expertise is often combined with random augmentation to pre-select subgrids for search, which severely limits the practicality of this approach as an autonomous solution. Summary of the invention

[0007] The following presents a summary of the invention to provide a basic understanding of one or more embodiments of the present invention. The present summary is not intended to identify key or important elements, nor is it intended to delineate any scope of a specific embodiment or any scope of the claims. Its sole purpose is to present concepts in a simplified form as a preface to a more detailed description presented later. In one or more embodiments described herein, a system, computer-implemented method, apparatus, and / or computer program product for an automated data enhancement process that may consider employing a random one-dimensional enhancement algorithm is described.

[0008] According to an embodiment, a system is provided. The system may include a processor operatively coupled to a memory that executes a computer executable component stored in the memory. The computer executable component may include a data enhancement component that may execute a random one-dimensional enhancement algorithm to enhance a data set for training a machine learning model via a plurality of enhancement operations. The random one-dimensional enhancement algorithm may employ a global enhancement parameter that defines: a distortion amplitude associated with a plurality of enhancement operations, and a number of enhancement operations included in the plurality of enhancement operations.

[0009] According to another embodiment, a computer-implemented method is provided. The computer-implemented method may include executing, by a system operably coupled to a processor, a random one-dimensional enhancement algorithm to enhance a data set used to train a machine learning model via a plurality of enhancement operations. The random one-dimensional enhancement algorithm may employ a global enhancement parameter that defines: a distortion magnitude associated with the plurality of enhancement operations, and a number of enhancement operations included in the plurality of enhancement operations.

[0010] According to another embodiment, a computer program product for data enhancement is provided. The computer program product may include a computer-readable storage medium having program instructions embodied therein. The program instructions may be executed by a processor to cause the processor to perform or be executed by the processor to perform a random one-dimensional enhancement algorithm to enhance a data set for training a machine learning model via a plurality of enhancement operations. The random one-dimensional enhancement algorithm may employ a global enhancement parameter that defines: a distortion amplitude associated with a plurality of enhancement operations, and the number of enhancement operations included in the plurality of enhancement operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A block diagram of an exemplary non-limiting system according to one or more embodiments described herein is shown that can generate a one-dimensional search space that can be used to train one or more machine learning models via autonomous random one-dimensional data augmentation.

[0012] Figure 2Figures showing exemplary non-limiting graphs that may depict machine learning model accuracy as a function of two global parameters according to one or more embodiments described herein.

[0013] Figure 3 A diagram illustrating an exemplary non-limiting graph that may depict preprocessing and training speed as a function of augmentation parameters according to one or more embodiments described herein.

[0014] Figure 4 A diagram of an exemplary non-limiting system according to one or more embodiments described herein is shown that can generate one or more transformation operations that can be used to train one or more machine learning models via autonomous random one-dimensional data augmentation.

[0015] Figure 5 A diagram showing an exemplary non-limiting table that may illustrate one or more transformation operations that may be generated for training one or more machine learning models via autonomous random one-dimensional data augmentation according to one or more embodiments described herein.

[0016] Figure 6 A diagram is shown of an exemplary non-limiting system that can execute an automated search algorithm for training one or more machine learning models via autonomous random one-dimensional data augmentation according to one or more embodiments described herein.

[0017] Figures 7 and 8 Figures showing exemplary non-limiting graphs that may demonstrate unimodal relationships characterizing one or more machine learning models according to one or more embodiments described herein.

[0018] Fig. 9 A diagram of exemplary non-limiting pseudocode according to one or more embodiments described herein is shown, which pseudocode may characterize one or more automatic search algorithms that may be used to train one or more machine learning models via autonomous random one-dimensional data augmentation.

[0019] Fig.10 A diagram showing an exemplary non-limiting table that can demonstrate the efficacy of generating one or more transformation operations for training one or more machine learning models via autonomous random one-dimensional data augmentation according to one or more embodiments described herein.

[0020] Fig.11 Figures showing exemplary non-limiting tables that may demonstrate the efficacy of one or more autonomous random one-dimensional data augmentations for training one or more machine learning models according to one or more embodiments described herein.

[0021] Fig.12 A flowchart is shown of an exemplary non-limiting computer-implemented method that may employ one or more autonomous random one-dimensional data augmentations for training one or more machine learning models according to one or more embodiments described herein.

[0022] Fig.13 A block diagram is shown of an exemplary non-limiting operating environment in which one or more embodiments described herein may be facilitated. DETAILED DESCRIPTION

[0023] The following specific embodiments are merely exemplary and are not intended to limit the application or use of the embodiments and / or the embodiments. In addition, it is not intended to be bound by any express or implied information set forth in the aforementioned "background technology" or "content of the invention" section or "specific embodiments" section.

[0024] One or more embodiments are now described with reference to the accompanying drawings, wherein the same reference numerals are used to represent the same elements throughout. In the following description, for the purpose of explanation, many specific details are set forth in order to provide a more thorough understanding of one or more embodiments. However, it is apparent that in various cases, one or more embodiments may be practiced without these specific details.

[0025] Considering the problems of other specific implementations that employ training data set augmentation; the present disclosure can be implemented to generate solutions to one or more of these problems by utilizing a single global augmentation parameter that can generate a one-dimensional search space. Advantageously, one or more embodiments described herein can perform one or more autonomous search algorithms in conjunction with a reduced search space to eliminate the need for human intervention (e.g., eliminating the need for subject matter experts to define relevant subgrids of the search space) and reduce computing resources.

[0026] Various embodiments of the present invention may be directed to computer processing systems, computer-implemented methods, apparatus, and / or computer program products that facilitate efficient, effective, and autonomous (e.g., without direct human guidance) data augmentation to improve deep learning model performance. For example, one or more embodiments described herein may employ a random one-dimensional augmentation ("RUA") process to employ multiple augmentation operations directly related to a single global augmentation parameter. Additionally, one or more embodiments described herein may reduce the training data search space to 10 1Technical improvements over conventional training dataset augmentation may include increasing the size of the training dataset, increasing the diversity of the transform domain, and / or alleviating the need for human expertise through efficient, autonomous search algorithms. In addition, one or more embodiments described herein may have practical applications by improving the performance of machine learning models. For example, various embodiments described herein may control the composition of the training dataset for a model, thereby affecting one or more model performance metrics that are independent of typical computational costs at inference time.

[0027] As used herein, the term "machine learning task" may refer to the application of artificial intelligence ("AI") techniques to automatically and / or autonomously learn and / or improve from experience (e.g., training data) without explicit programming of the lessons learned and / or improved. For example, a machine learning task may utilize one or more algorithms to facilitate supervised and / or unsupervised learning to perform tasks such as classification, regression, and / or clustering. According to one or more model configuration settings, the performance of the machine learning task may be facilitated by one or more machine learning models trained on one or more data sets.

[0028] As used herein, the term "machine learning model" may refer to a computer model that can be used to facilitate one or more machine learning tasks. For example, a computer model may simulate a plurality of interconnected processing units that may be similar to an abstract version of a neuron. For example, the processing units may be arranged in a plurality of layers (e.g., one or more input layers, one or more hidden layers, and / or one or more output layers) connected by varying connection strengths (e.g., commonly referred to as "weights" in the art). Machine learning can be learned through training, wherein data with known outcomes are input into a computer model, outputs about the data are compared to the known outcomes, and / or the weights of the computer model are autonomously adjusted based on repeated comparisons of known outcomes. As used herein, the term "training data set" may refer to data and / or data sets used to train one or more neural network models. As a machine learning model is trained (e.g., with more training data sets), the machine learning model may become increasingly accurate; therefore, a trained machine learning model may accurately analyze data with unknown outcomes based on the lessons learned from the training data to facilitate one or more machine learning tasks. Exemplary machine learning models may include, but are not limited to: Perceptron (“P”), Feedforward (“FF”), Radial Basis Network (“RBF”), Deep Feedforward (“DFF”), Recurrent Neural Network (“RNN”), Long / Short Term Memory (“LSTM”), Gated Recurrent Unit (“GRU”), Autoencoder (“AE”), Gated Recurrent Unit (“GRU”), Autoencoder (“AE”), Variational AE (“VAE”), Denoising AE (“DAE”), Sparse AE (“SAE”), Markov Chain (“MC”), Hopf ield network (“HN”), Boltzmann machine (“BM”), deep belief network (“DN”), deep convolutional network (“DCN”), deconvolutional network (“DN”), deep convolutional inverse graphics network (“DCIGN”), generative adversarial network (“GAN”), liquid state machine (“LSM”), extreme learning machine (“ELM”), echo state network (“ESN”), deep residual network (“DRN”), kohonen network (“KN”), support vector machine (“SVM”), and / or neural Turing machine (“NTM”).

[0029] Figure 1A block diagram of an exemplary non-limiting system 100 that can autonomously enhance one or more training data sets of one or more machine learning models is shown. For the sake of brevity, repeated descriptions of similar elements used in other embodiments described herein are omitted. Aspects of the systems (e.g., system 100, etc.), devices, or processes in various embodiments of the present invention may constitute one or more machine-executable components embodied in one or more machines (e.g., embodied in one or more computer-readable media associated with one or more machines). When executed by one or more machines (e.g., computers, computing devices, virtual machines, combinations thereof, etc.), such components may cause the machines to perform the operations described.

[0030] like Figure 1 As shown, system 100 may include one or more servers 102, networks 104, and / or input devices 106. Server 102 may include a data enhancement component 110. Data enhancement component 110 may also include a communication component 112 and / or a global parameter component 114. In addition, server 102 may include or otherwise be associated with at least one memory 116. Server 102 may also include a system bus 118, which may be coupled to various components such as, but not limited to, data enhancement component 110 and associated components, memory 116, and / or processor 120. Although Figure 1 Server 102 is shown in FIG. 1 , but in other embodiments, multiple devices of various types may be associated with Figure 1 The features shown in are associated with or include these features. In addition, the server 102 can communicate with one or more cloud computing environments.

[0031] The one or more networks 104 may include wired and wireless networks, including, but not limited to, cellular networks, wide area networks (WANs) (e.g., the Internet), or local area networks (LANs). For example, the server 102 may communicate with the one or more input devices 106 (or vice versa) using virtually any desired wired or wireless technology, including, but not limited to, cellular, WAN, wireless fidelity (Wi-Fi), Wi-Max, WLAN, Bluetooth technology, combinations thereof, and the like. Furthermore, while in the illustrated embodiment, the data enhancement component 110 may be provided on one or more servers 102, it should be understood that the architecture of the system 100 is not limited thereto. For example, the data enhancement component 110 or one or more components of the data enhancement component 110 may be located on another computer device, such as another server device, a client device, and the like.

[0032] The one or more input devices 106 may include one or more computerized devices, which may include, but are not limited to, personal computers, desktop computers, laptop computers, cellular phones (e.g., smart phones), computerized tablets (e.g., including processors), smart watches, keyboards, touch screens, mice, combinations thereof, etc. In various embodiments, the one or more input devices 106 may be used to input one or more settings 122 (e.g., one or more enhancement operations, global enhancement parameters, and / or machine learning tasks may be defined according to various embodiments described herein), training data sets 124, and / or machine learning models 126 into the system 100, thereby sharing the data with the server 102 (e.g., via a direct connection and / or via one or more networks 104). For example, the one or more input devices 106 may send data to the communication component 112 (e.g., via a direct connection and / or via one or more networks 104). In addition, the one or more input devices 106 may include one or more displays that may present one or more outputs generated by the system 100 to a user. For example, one or more displays may include, but are not limited to, cathode ray tube displays ("CRT"), light emitting diode displays ("LED"), electroluminescent displays ("ELD"), plasma display panels ("PDP"), liquid crystal displays ("LCD"), organic light emitting diode displays ("OLED"), combinations thereof, and the like.

[0033] In various embodiments, one or more input devices 106 and / or one or more networks 104 may be used to input one or more settings 122 and / or commands into the system 100. For example, in various embodiments described herein, one or more input devices 106 may be used to operate and / or manipulate the server 102 and / or associated components. In addition, one or more input devices 106 may be used to display one or more outputs (e.g., displays, data, visualizations, etc.) generated by the server 102 and / or associated components. In addition, in one or more embodiments, one or more input devices 106 may be included in a cloud computing environment and / or may be operably coupled to a cloud computing environment. In various embodiments, one or more input devices 106 may be used to input one or more training data sets 124 into the system 100 for enhancement by the data enhancement component 110. The enhanced training data sets 124 may thus be used to train one or more machine learning models 126. In various embodiments, the enhancement component may share one or more enhanced training data sets 124 with one or more input devices 106, and / or directly supply one or more enhanced training data sets 124 to one or more machine learning models 126 to facilitate training.

[0034] In one or more embodiments, the data enhancement component 110 can execute the RUA algorithm to enhance one or more training data sets 124 for training the machine learning model 126. According to various embodiments described herein, the one or more training data sets 124 can be supplied via one or more input devices 106. Alternatively or in addition, one or more of the training data sets 124 can be retrieved from one or more memories 116 of the system 100. For example, the data enhancement component 110 (e.g., via the communication component 112) can retrieve one or more training data sets 124 from one or more memories 116 used as a data set repository.

[0035] In various embodiments, the RUA algorithm may perform one or more augmentation operations on one or more training data sets 124. For example, the data augmentation component 110 may select an augmentation operation to be employed by the RUA algorithm from a pool of possible augmentation operations 128. In one or more embodiments, the one or more machine learning models 126 may be trained via multiple training steps. With each training step, the data augmentation component 110 may augment one or more training data sets 124 to generate one or more new augmented training data sets 124. Thus, each training step may be associated with a corresponding selection of an augmentation operation by the data augmentation component 110. In various embodiments, the data augmentation component 110 may randomly (e.g., with uniform probability, with replacement) select an augmentation operation.

[0036] like Figure 1As shown, the pool of possible enhancement operations 128 may be stored in one or more memories 116. In various embodiments, the one or more enhancement operations included in the pool of possible enhancement operations 128 may be provided as one or more settings 122 input into the system via one or more input devices 106. In addition, the one or more enhancement operations included in the pool of possible enhancement operations 128 may be predefined and / or may be operations utilized by the data enhancement component 110 in previous training data set 124 enhancements. In various embodiments, the composition of the pool of possible enhancement operations 128 may depend on one or more characteristics of one or more training data sets 124. For example, the composition of the pool of possible enhancement operations 128 may depend on the type of data included in one or more training data sets 124, the amount of data, and / or the size of the data. In one or more embodiments, the training data set 124 may be associated with a corresponding pool of possible enhancement operations 128. Exemplary enhancement operations that may be included in one or more pools of possible enhancement operations 128 may include, but are not limited to, contrast adjustment operations, histogram equalization operations, rotation operations, overexposure operations, tone separation operations, color adjustment operations, brightness adjustment operations, sharpness adjustment operations, one or more shearing operations (e.g., with respect to the X-axis and / or Y-axis), one or more translation operations (e.g., with respect to the X-axis and / or Y-axis), scaling operations, de-coordination operations, blurring operations, noise addition, grid distortion, combinations thereof, and the like.

[0037] In one or more embodiments, the global parameter component 114 may define a global enhancement parameter ("P") employed by the RUA algorithm. In various embodiments, the RUA algorithm may employ a single global enhancement parameter P to affect each enhancement operation included in one or more pools of possible enhancement operations 128. For example, the global parameter component 114 may set a single global enhancement parameter P to cover multiple parameters, including, but not limited to: a distortion amplitude parameter; multiple enhancement operation parameters, application probabilities, combinations thereof, and the like. For example, a single global enhancement parameter P may define: how many enhancement operations are selected from one or more pools of possible enhancement operations 128 for a given training iteration; and / or a distortion amplitude (e.g., which may control the amount of enhancement produced by implementing the corresponding enhancement operation). In one or more embodiments, the global parameter component 114 may query one or more input devices 106 to define the global enhancement parameter P. In one or more embodiments, the global parameter component 114 may set the global enhancement parameter P according to one or more predefined values. In one or more embodiments, the global enhancement parameter P may be within a predefined integer range (e.g., 1-10 or higher). In one or more embodiments, the global parameter component 114 defines a global enhancement parameter P based on multiple parameter values ​​defined in one or more settings 122 provided via one or more input devices 106 (e.g., the global enhancement parameter can be a function of multiple parameter values, such as a weighted average).

[0038] In various embodiments, the data enhancement component 110 may perform the RUA algorithm in multiple training iterations. For each training iteration, the data enhancement component 110 may select different values ​​of the global enhancement parameter P from a one-dimensional search space. In addition, with each training iteration, the data enhancement component 110 (e.g., and / or one or more associated components thereof according to various embodiments described herein) may measure the performance of the machine learning model 126 relative to one or more performance metrics (e.g., accuracy, precision, computing resources utilized, speed, combinations thereof, etc.). By exploring possible P values ​​in multiple training iterations, the data enhancement component 110 may determine the optimal value of the global enhancement parameter P. Therefore, by reducing the search space to a one-dimensional space, the data enhancement component 110 may determine the optimal P value with few training iterations and minimal computing resources.

[0039] In contrast, typical data augmentation strategies (e.g., such as random augmentation) perform augmentation operations based on multiple global parameters. As the number of global parameters increases, the resulting search space also increases. For example, considering the distortion magnitude (e.g., parameter "M") and the number of augmentation operations (e.g., parameter "N") as separate global parameters (e.g., rather than as a single global augmentation parameter according to various embodiments described herein) may yield a search space of size 10 2A search space of (e.g., an order of magnitude larger than the one-dimensional search space achievable according to the various embodiments described herein).

[0040] Furthermore, as the size of the search space increases, the number of training iterations required to determine the best augmentation setting may also increase. 2 A search space of can result in a 10×10 grid search, which may require 100 iterations of training to find the best augmentation setting. To avoid the computational cost associated with 100 training iterations, a subgrid is often selected before performing the search. However, the appropriate subgrid selection is highly dependent on the network and / or dataset involved, reintroducing the need for human expertise and / or experience.

[0041] By defining a single global enhancement parameter to be employed by one or more RUA algorithms, the global parameter component 114 can reduce the dimensionality of the search space (e.g., compared to a typical enhancement protocol that can employ multiple global parameters). Thus, the data enhancement component 110 can reduce the search space, optimize the enhancement parameters, and employ one or more efficient search algorithms to negate human intervention while minimizing computational resources according to various embodiments described herein.

[0042] Figure 2 Figures of exemplary non-limiting graphs 202, 204 are shown according to one or more embodiments described herein, which can demonstrate the efficacy of employing a single global enhancement parameter P (e.g., via a global parameter component 114). For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. To establish graphs 302 and / or 304, a random enhancement algorithm is performed on a full 10×10 grid for two machine learning tasks (e.g., classification tasks) using machine learning models 126 ResNet9 and WRN-28-2 for training data sets 124 Cifar10 and / or SVHN. Graphs 202 and / or 204 depict model accuracy as a function of the following two global parameters (e.g., processed as separate parameters): a distortion amplitude parameter M and the number of enhancement operations N.

[0043] like Figure 2As shown, the grayscale gradient in the graph 202 and / or 204 forms a diagonal line from the upper left of the graph 202 and / or 204 to the lower right of the graph 202 and / or 204. Although the accuracy value and / or the best point between the graphs 202 and 204 vary, both exhibit a gradient that approximates a diagonal line. The presence of the diagonal gradient indicates that the two global parameters can be merged (e.g., via the global parameter component 114) to reduce the search space along the diagonal gradient while still capturing the accuracy of the model configuration. In various embodiments, the global parameter component 114 can merge multiple parameters to form a global enhancement parameter P by linear combination. In addition, the coefficients of the linear combination can be adjusted to take into account the performance of the computing resources and / or the machine learning model 126.

[0044] Figure 3 A diagram of an exemplary non-limiting graph 302 is shown that can display preprocessing and training speed as a function of training iterations according to one or more embodiments described herein. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. In various embodiments, the parameters covered by the global enhancement parameter P can have different degrees of impact on one or more performance metrics of one or more machine learning models 126. Therefore, the global parameter component 114 can define the global enhancement parameter P so that the value of the global enhancement parameter P can be converted to corresponding values ​​(e.g., different values) of the covered parameters.

[0045] For example, Figure 3 It is shown that in the case where the global enhancement parameter P covers the distortion magnitude M and the number of enhancement operations N, the adjustment of the number of enhancement operations N can have a greater impact on the preprocessing and training speed of the machine learning model 126 than the adjustment of the distortion magnitude M. To create the graph 302, data augmentation was applied to the machine learning model 126 ResNet9 using the training data set 124 Cifar10 using two global parameters M and N. Line 304 represents the preprocessing speed, and line 306 represents the training speed.

[0046] like Figure 3 As shown, applying a large number of augmentations during training can severely limit the training speed. For example, graph 302 illustrates that when the number of augmentation operations N is greater than or equal to 3 (e.g., N ≥ 3), central processor unit (“CPU”) based preprocessing can become rate limiting (e.g., especially when N ≥ 5). In various embodiments, global parameter component 114 can define a global augmentation parameter P such that based on Figure 3 According to the preprocessing and training speed relationship depicted in , the maximum number of enhancement operations is capped at 5. For example, the global parameter component 114 can set the number of enhancement operations associated with the global enhancement parameter P according to the following equation 1.

[0047]

[0048] In various embodiments, when defining the global augmentation parameter P, the maximum number of augmentation operations may be set to a value greater than 5. For example, the global parameter component 114 may define the global augmentation parameter P based on computing resources available for training one or more machine learning models 126 (e.g., the computing resources may be included in one or more settings 122 defined by one or more input devices 106).

[0049] Figure 4 A diagram of an exemplary non-limiting system 100 according to one or more embodiments described herein is shown, the system also including a parameter component 402. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. In one or more embodiments, the parameter component 402 can generate parameter definitions 404 for enhancement operations included in one or more pools of possible enhancement operations 128. For example, the parameter component 402 can generate a parameter definition 404 that aligns (e.g., is directly related to) one or more enhancement operations with a global enhancement parameter P. Figure 4 As shown, one or more parameter definitions 404 can be stored in one or more memories 116 for subsequent retrieval by the data enhancement component 110 and / or its associated components.

[0050] In various embodiments, the search space can be reduced to a one-dimensional space using a single global enhancement parameter P. In addition, the global enhancement parameter P can control the enhancement strength (e.g., the amount of enhancement) associated with one or more enhancement operations via direct association. For example, a zero value of the global enhancement parameter P can result in no enhancement, while increasing the value of the global enhancement parameter P can result in an increased amount of enhancement. The parameter component 402 can generate a parameter definition 404 associated with the corresponding enhancement operation so that P=0 results in no enhancement. In addition, the parameter component 402 can define a maximum enhancement strength for one or more enhancement operations. In addition, in one or more embodiments, the parameter component 402 can apply a random uniform distribution ("U") to the parameter definition 404 of the enhancement operation to control the density of the search space (e.g., about the enhancement operation).

[0051] In one or more embodiments, one or more input devices 106 may be used to input one or more enhancement operations into and / or associate a definition to the system 100. In addition, the parameter component 402 may check whether the provided definition is aligned with the global enhancement parameter P. For example, the parameter component 402 may determine whether an increase in the value of the global enhancement parameter P also results in an increase in the enhancement achieved by a given enhancement operation according to a given definition. In the case where an increase in the value of the global enhancement parameter P may result in a reduction in the amount of enhancement achieved by the enhancement operation, the parameter component 402 may determine that the given definition is not aligned with the global enhancement parameter P. For example, a given definition may define an enhancement operation as the inverse of a parameter covered by the global enhancement parameter P. In the case where a given definition is not aligned with the global enhancement parameter P, the parameter component 402 may adjust a given parameter to generate a parameter definition 404 associated with a given enhancement operation, which is aligned with the global enhancement parameter P. For example, the parameter component 402 may transform an inverse correlation in a given definition into a direct correlation in a parameter definition 404. In addition, the parameter component 402 may generate one or more parameter definitions 404 that align the off-center correlation of a given definition with a central linear correlation.

[0052] Figure 5 A diagram of an exemplary non-limiting table 500 generated according to one or more embodiments described herein is shown, which can display exemplary parameter definitions 404 that can be generated by parameter component 402. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. Table 500 can depict exemplary parameter definitions 404 associated with enhancement operations of an exemplary pool of possible enhancement operations 128, from which data enhancement component 110 can select. In addition, table 500 can depict exemplary parameter definitions 404 that can be generated by parameter component 402 compared to definitions used by typical enhancement strategies (e.g., random enhancement). In Figure 5 For random enhancement, r = M / M max , and for the parameter definition 404 generated by the parameter component 402, r = P / P max . Boosting operations marked with an asterisk ("*") may have non-zero impact at r=0 under random boosting (e.g., may not be aligned with the global boosting parameter P), but are zero-aligned according to the parameter definition 404 generated by the parameter component 402 and / or used thereby in one or more RUA algorithms.

[0053] Although table 500 depicts 14 exemplary enhancement operations and associated definitions (e.g., parameter definitions 404), the architecture of the RUA algorithm is not limited thereto. For example, embodiments are also contemplated where the size of the pool of possible enhancement operations 128 and / or associated parameter definitions 404 is greater or less than 14. For example, in one or more embodiments, parameter component 502 can generate parameter definitions 404 for additional enhancement operations, including, but not limited to: scaling, decoordination, blurring, combinations thereof, and the like.

[0054] like Figure 5 As shown, the boosting operation of a typical boosting strategy (e.g., random boosting) cannot be scaled with a single global parameter (e.g., and thus cannot be used in a one-dimensional search space, such as the space established by the global parameter component 114). Figure 5 The enhancement strength of the overexposure enhancement operation and the tone separation enhancement operation shown is inversely proportional to the parameter M of the random enhancement. In addition, Figure 5 The color enhancement operation, contrast enhancement operation, brightness enhancement operation and / or sharpness enhancement operation illustrated in the example are shifted because when , they result in no enhancement, while values ​​closer to 0 or greater than 1 result in stronger changes to the input.

[0055] In addition, typical augmentation strategies (e.g., random augmentation) use deterministic augmentation for any given global parameter M. For example, when M=M max , whenever a rotation augmentation operation is applied, the image will be rotated by ±30 degrees. The deterministic behavior significantly reduces the diversity in the augmentation space, and thus can allow the machine learning model 126 to overfit more easily. In contrast, the parameter component 402 can generate a parameter definition such that P=0 will result in no augmentation. In addition, the parameter component 402 can extend the maximum rotation degree from ±30 degrees to ±90 degrees.

[0056] Figure 6 A diagram of an exemplary, non-limiting system 100 according to one or more embodiments described herein is shown, the system also including a search component 602. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. In various embodiments, the search component 602 can employ one or more search algorithms that are more efficient than a grid search to explore a search space with minimal computing resources.

[0057] Graphs 202 and / or 204 further illustrate that when traversing the diagonal gradient, the accuracy of the machine learning model 126 may first increase to a maximum value and then decrease. Thus, the machine learning model 126 may experience a single peak with respect to a given parameter (such as the distortion amplitude parameter M). Figure 7A diagram of exemplary non-limiting graphs 702 and / or 704 according to one or more embodiments described herein that may illustrate a single peak of one or more machine learning models 126 is shown. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. To create graphs 702 and / or 704, diagonal terms from graphs 202 and / or 204 are extracted and the associated relative accuracy is plotted against parameter M. Thus, graph 702 may depict a single peak inherent to graph 202, and graph 704 may depict a single peak inherent to graph 204.

[0058] in addition, Figure 8 800 are shown, which may further verify the single peak of one or more machine learning models 126 according to one or more embodiments described herein. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. To create the graph 800, the parameter definition 404 generated by the parameter component 402 and illustrated in the table 500 may be used. max = 30 is applied, and then the relative performance is plotted against the global enhancement parameter P. Regarding the graph 800, the data enhancement component 110 uses Figure 5 The exemplary pool of augmentation operations 128 and parameter definitions 404 shown are used to perform the RUA algorithm to train the machine learning model 126 PyramidNet for the training data set 124 Cifar10. The curve graph 800 verifies the unimodal relationship and shows that despite the small randomness at the local scale, the overall unimodal relationship between accuracy and the global augmentation parameter P can persist as a function of the global parameter component 114 and the parameter component 402. In various embodiments, the data augmentation component 110 can generate similar curve graphs for a given machine learning model 126 and / or training data set 124 to verify the unimodality.

[0059] Based on the unimodal characteristics of one or more machine learning models 126, the search component 602 may employ one or more search algorithms that can utilize unimodal functions. For example, the search component 602 may execute a golden section search algorithm that can find the maximum and / or minimum of a unimodal function on a given interval. Other exemplary search algorithms that may be employed by the search component 602 may include, but are not limited to: binary search, tree search, interpolation search, combinations thereof, and the like. By employing one or more search algorithms (e.g., via the search component 602) on a reduced search space (e.g., generated by a function of the global parameter component 114) with an optimized parameter definition 404 (e.g., via the parameter component 402), the data enhancement component 110 may explore search locations without a grid search guided by human expertise (e.g., as required by a typical enhancement strategy) and with fewer training iterations.

[0060] Fig. 9 FIG. 9 is a diagram showing an exemplary non-limiting pseudo code 900 of a golden section search algorithm according to one or more embodiments described herein that can be employed by the search component 602. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. Fig. 9 As shown, "f" may represent the function to be evaluated, "a" may represent the lower limit of the search space, "b" may represent the upper limit of the search space, "maxIter" may represent the maximum number of iterations to run the search, and the return value may be the domain value corresponding to the maximum evaluation of "f". Using the golden section search algorithm represented by pseudo code 900, each evaluation of the search space may reduce the remaining search space by a constant factor of approximately 0.618 (e.g., the inverse of the golden ratio). Therefore, after 6 to 7 evaluations from the initial search space of size 30, the data enhancement component 110 may solve for the integer value P. In other words, the data enhancement component 110 may repeat the training task 6 to 7 times to find the optimal global enhancement parameter P.

[0061] Fig.10 A diagram of an exemplary non-limiting table 1002 is shown, which can further demonstrate the efficacy of the data enhancement component 110 according to one or more embodiments described herein. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. For example, the table 1002 can be established from an ablation study of parameter definitions 404, which can be generated by the parameter component 402 and illustrated in the table 500. Specifically, the ResNet9 machine learning model 126 is trained on the Cifar10 training data set 124, where the accuracy is averaged over 10 independent runs.

[0062] As described herein, parameter component 402 may generate parameter definitions 404 having an impact that is positively correlated with P, may employ a random uniform distribution, and may extend one or more enhancement parameters. Table 1002 depicts 8 rows, each row corresponding to a respective combination of possible functions (e.g., "aligned," "random," "extended") of parameter component 402 when generating parameter definitions 404, where "0" indicates that the function was not employed and "1" indicates that the function was employed. For example, the "aligned" column may indicate whether parameter component 402 generated parameter definitions 404 such that an enhancement operation is directly related to a global enhancement parameter P (e.g., an increase in the value of the global enhancement parameter P translates into an increase in the amount of enhancement achieved by the enhancement operation). The "random" column may indicate whether parameter component 402 generated parameter definitions 404 having a random uniform distribution. The "extended" column may indicate whether parameter component 402 extended one or more parameter ranges when generating parameter definitions 404 (e.g., as compared to a typical enhancement implementation), such as Figure 5At least the rotation enhancement operation shown is illustrated.

[0063] like Fig.10 As shown, row 1 of table 1002 may be similar to the one in which the random boosting operation and definition (e.g., defined in table 500) are combined with the golden section search algorithm (e.g., where P max =30). Row 8 may represent an implementation in which parameter component 402 is combined with a golden section search algorithm (eg, where P max =30) employs an implementation in which all three functions are employed in generating parameter definitions 404 (eg, as illustrated in table 500). Rows 2 through 7 may represent implementations in which parameter definitions 404 are generated using various combinations of the three functions of parameter component 402.

[0064] As shown in table 1002, correcting the alignment of the enhancements marked with "*" from table 500 so that their impact is positively correlated with P may be beneficial; as demonstrated by comparison of rows: 1 vs. 5, 2 vs. 6, 3 vs. 7, and 4 vs. 8. Additionally, employing a random uniform distribution when generating parameter definitions 404 may also be beneficial; as demonstrated by comparison of rows: 1 vs. 3, 2 vs. 4, 5 vs. 7, and 6 vs. 8. Furthermore, increasing the maximum strength of the enhancement (e.g., employing a rotation of the enhancement operation by ±90 instead of ±30) may be detrimental in itself, but is beneficial when paired with random sampling (e.g., as demonstrated by comparison of rows 3 vs. 4 and / or 7 vs. 8).

[0065] Fig.11An exemplary non-limiting table 1102 and / or a diagram of 1104 according to one or more embodiments described herein are shown to further demonstrate the efficacy of the data enhancement component 110 relative to a typical enhancement strategy. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. Table 1102 depicts the experimental parameter details associated with implementing the data enhancement component 110 to train the following publicly available machine learning model 126 networks: PyramidNet-272-200, ShakeDrop (0.5), Wide-ResNet-28-10, Wide-ResNet-28-10, Wide-ResNew-28-2. In addition, as shown in Table 1102, the following publicly available training data sets 124 are utilized: Cifar10, Cifar100, and SVHN. As shown in Table 1102, mean and / or standard deviation data normalization ("mean-standard deviation-normalization") may be utilized. In addition, additional enhancement algorithms (e.g., "Padding Crop," "Horizontal Flip," and / or "Cut") may be utilized before and / or after executing the RUA algorithm. In addition, a stochastic gradient descent ("SGD") optimizer may be utilized. In addition, "LR" in Table 1102 may represent a learning rate. In each training data set 124, five thousand training samples are retained as evaluation data for selecting the best parameter P. After P is selected, the retained set is merged back into the training data set 124, and the machine learning model 126 is trained again. The test performance at the end of the final training iteration is then recorded.

[0066] Table 1104 depicts the final test results of the RUA algorithm performed by the data augmentation component 110 compared to typical augmentation strategies (e.g., random augmentation "RA", automatic augmentation "AA", fast automatic augmentation "FastAA" and / or population-based augmentation "PBA"), where the performance scores are averaged from 10 independent runs. The best accuracy for each column from Table 1104 is Fig.11 1104, the RUA algorithm as performed by the data augmentation component 110 achieves equal or better test scores (i.e., typical augmentation policies for 3 of the 4 tasks) while reducing the search space by an order of magnitude. For each of the 3 tasks, the improvement over the runner-up augmentation policy is statistically significant, with one-tailed t-test p-values ​​of 0.003, 0.001, and 0.011. On the dataset SVHN, the RUA algorithm as performed by the data augmentation component 110 achieves competitive performance.

[0067] Fig.12A flow chart of an exemplary non-limiting computer-implemented method 1200 is shown that can facilitate augmenting one or more training data sets 124 with one or more RUA algorithms according to one or more embodiments described herein. For the sake of brevity, repeated descriptions of similar elements employed in other embodiments described herein are omitted. Fig.12 An example order of events is shown, but various features of the computer-implemented method 1200 may be performed in alternate orders.

[0068] At 1202, the computer-implemented method 1200 may include receiving (e.g., via the communication component 112) by the system 100 operatively coupled to the processor 120 one or more training data sets 124. In various embodiments, the one or more training data sets 124 may be used to train one or more machine learning models 126. According to various embodiments described herein, one or more settings 122 and / or machine learning models 126 may also be received at 1202. Additionally, one or more user preferences may be included in the data received at 1202, such as preferred enhancement operations and / or parameter settings.

[0069] At 1204, the computer-implemented method 1200 may include generating (e.g., via the global parameter component 114) a one-dimensional search space by the system based on a global enhancement parameter P and / or a plurality of enhancement operations selected from a pool of possible enhancement operations 128. According to various embodiments described herein, generating the one-dimensional search space may include defining a global enhancement parameter P. For example, the global enhancement parameter P may be defined to encompass a plurality of parameters associated with the enhancement operation. For example, the global enhancement parameter P may characterize both the distortion magnitude and the number of transformations associated with the enhancement training data set. In various embodiments, the computer-implemented method 1300 may utilize a single global enhancement parameter to control the enhancement of one or more training data sets 124.

[0070] At 1206, the computer-implemented method 1200 may include generating, by the system 100 (e.g., via the parameter component 402), parameter definitions 404 for the pool of possible enhancement operations 128. According to various embodiments described herein, the parameter definitions 404 may be generated to: directly relate to the global enhancement parameter P, incorporate a random uniform distribution, and / or extend one or more given parameter ranges. For example, table 500 illustrates a plurality of parameter definitions 404 that may be generated in association with 14 exemplary enhancement operations.

[0071] At 1208, the computer-implemented method 1300 may include executing (e.g., via the search component 702) one or more search algorithms by the system 100 on the one-dimensional search space to select a new value of the global enhancement parameter P for a given training iteration. For example, the search algorithm executed at 1208 may exploit the unimodal relationship exhibited by the machine learning model 126 to be trained. For example, the search algorithm may be a golden section search algorithm (e.g., such as Fig. 9 exemplified). In various embodiments, the computer-implemented method 1200 may execute the search algorithm in multiple training iterations. For example, at 1210, the computer-implemented method 1200 may include determining (e.g., via the data enhancement component 110) whether the entire search space has been explored by the search algorithm. In the event that the entire search space has not been explored, the computer-implemented method 1200 may return to 1208. In the event that the entire search space has been explored, the computer-implemented method 1200 may proceed to 1212. In one or more embodiments, the determination at 1210 may be made with reference to a defined exploration threshold (e.g., defined via one or more input devices 106) rather than with reference to the entire search space.

[0072] At 1212, the computer-implemented method 1200 can include identifying (e.g., via the search component 602) by the system 100 an optimal global augmentation parameter P value from performance data characterizing the performed training iterations. For example, performance data regarding the machine learning model 126 can be collected in association with each new global augmentation parameter P value selected at 1208. Based on the performance data for all global augmentation parameter P values ​​selected by performing 1208 one or more times, an optimal global augmentation parameter P value can be identified.

[0073] To provide additional context for the various embodiments described herein, Fig.13 The following discussion is intended to provide a general description of a suitable computing environment 1300 in which various embodiments of the embodiments described herein may be implemented. Although the embodiments have been described above in the general context of computer-executable instructions that may be executed on one or more computers, those skilled in the art will recognize that the embodiments may also be implemented in conjunction with other program modules and / or as a combination of hardware and software.

[0074] Generally, program modules include routines, programs, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In addition, those skilled in the art will appreciate that the methods of the present invention can be practiced with other computer system configurations, including single-processor or multi-processor computer systems, minicomputers, mainframe computers, Internet of Things (IoT) devices, distributed computing systems, as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, etc., each of which can be operatively coupled to one or more associated devices.

[0075] The illustrated embodiments of the embodiments herein can also be put into practice in a distributed computing environment, in which a specific task is performed by a remote processing device linked by a communication network. In a distributed computing environment, program modules can be located in local and remote memory storage devices. For example, in one or more embodiments, a computer executable component can be executed from a memory that may include one or more distributed memory units or be composed of one or more distributed memory units. As used herein, the terms "memory" and "memory unit" are interchangeable. In addition, one or more embodiments described herein can execute the code of the computer executable component in a distributed manner, for example, a plurality of processors are combined or work in collaboration to execute the code from one or more distributed memory units. As used herein, the term "memory" can cover a single memory or memory unit at a location, or a plurality of memories or memory units at one or more locations.

[0076] Computing devices typically include various media, which may include computer-readable storage media, machine-readable storage media, and / or communication media, where the two terms are used differently in this article, as described below. Computer-readable storage media or machine-readable storage media can be any available storage media that can be accessed by a computer, and include volatile and non-volatile media, removable and non-removable media. By way of example and not limitation, computer-readable storage media or machine-readable storage media can be implemented in conjunction with any method or technology for storing information such as computer-readable or machine-readable instructions, program modules, structured data, or unstructured data.

[0077] Computer-readable storage media may include, but are not limited to, random access memory ("RAM"), read-only memory ("ROM"), electrically erasable programmable read-only memory ("EEPROM"), flash memory or other memory technology, compact disk read-only memory ("CD-ROM"), digital versatile disk ("DVD"), Blu-ray disk ("BD") or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, solid-state drives or other solid-state storage devices, or other tangible and / or non-transitory media that can be used to store the desired information. In this regard, the terms "tangible" or "non-transitory" as applied to storage, memory, or computer-readable media herein should be understood to exclude only propagating transient signals per se as a modifier, and not to disclaim all standard storage, memory, or computer-readable media that are not merely propagating transient signals per se.

[0078] Computer-readable storage media may be accessed by one or more local or remote computing devices, eg, through access requests, queries, or other data retrieval protocols, to perform various operations on the information stored by the media.

[0079] Communication media typically embodies computer readable instructions, data structures, program modules, or other structured or unstructured data in a data signal, which may be a modulated data signal such as a carrier wave or other transport mechanism, and includes any information delivery or transmission media. The term "modulated data signal" or "signal" refers to a signal that has one or more of its characteristics set or changed to encode information in one or more signals. By way of example, and not limitation, communication media include wired media, such as a wired network or direct-wired connection, and wireless media, such as acoustic, RF, infrared and other wireless media.

[0080] Reference again Fig.13 , an exemplary environment 1300 for implementing various embodiments of aspects described herein includes a computer 1302, which includes a processing unit 1304, a system memory 1306, and a system bus 1308. The system bus 1308 couples system components including, but not limited to, the system memory 1306 to the processing unit 1304. The processing unit 1304 can be any of a variety of commercially available processors. Dual microprocessors and other multi-processor architectures can also be used as the processing unit 1304.

[0081] The system bus 1308 may be any of several types of bus structures that may further interconnect to a memory bus (with or without a memory controller), a peripheral bus, and a local bus using any of a variety of commercially available bus architectures. The system memory 1306 includes ROM 1310 and RAM 1312. A basic input / output system ("BIOS") may be stored in a non-volatile memory such as ROM, an erasable programmable read-only memory ("EPROM"), EEPROM, where the BIOS contains the basic routines that help transfer information between elements within the computer 1302, such as during startup. The RAM 1312 may also include high-speed RAM, such as static RAM for caching data.

[0082] The computer 1302 also includes an internal hard disk drive ("HDD") 1314 (e.g., EIDE, SATA), one or more external storage devices 1316 (e.g., a magnetic floppy disk drive ("FDD") 1316, a memory stick or flash drive reader, a memory card reader, a combination thereof, etc.), and an optical drive 1320 (e.g., which can read or write from a disk 1322 such as a CD-ROM disk, a DVD, a BD, etc.). Although the internal HDD 1314 is shown as being located within the computer 1302, the internal HDD 1314 can also be configured for external use in a suitable infrastructure (not shown). In addition, although not shown in the environment 1300, a solid-state drive ("SSD") can be used in addition to or instead of the HDD 1314. The HDD 1314, the external storage device 1316, and the optical drive 1320 can be connected to the system bus 1308 via an HDD interface 1324, an external storage interface 1326, and an optical drive interface 1328, respectively. Interface 1324 for external drive implementations may include at least one or both of Universal Serial Bus ("USB") and Institute of Electrical and Electronics Engineers ("IEEE") 1394 interface technologies. Other external drive connection technologies are within the contemplation of the embodiments described herein.

[0083] The drives and their associated computer-readable storage media provide non-volatile storage of data, data structures, computer-executable instructions, etc. For the computer 1302, the drives and storage media accommodate the storage of any data in a suitable digital format. Although the above description of computer-readable storage media refers to corresponding types of storage devices, it should be understood by those skilled in the art that other types of computer-readable storage media (whether currently existing or developed in the future) can also be used in the exemplary operating environment, and further, any such storage media can contain computer-executable instructions for performing the methods described herein.

[0084] A number of program modules may be stored in the drives and RAM 1312, including an operating system 1330, one or more application programs 1332, other program modules 1334, and program data 1336. All or portions of the operating system, application programs, modules, and / or data may also be cached in RAM 1312. The systems and methods described herein may be implemented using various commercially available operating systems or combinations of operating systems.

[0085] Computer 1302 may optionally include emulation technology. For example, a hypervisor (not shown) or other intermediary may emulate the hardware environment for operating system 1330, and the emulated hardware may optionally be different from the hardware of the operating system 1330. Fig.13 1302. In such embodiments, operating system 1330 may include one of a plurality of virtual machines ("VMs") hosted at computer 1302. In addition, operating system 1330 may provide a runtime environment, such as a Java runtime environment or a .NET framework, for application 1332. A runtime environment is a conforming execution environment that allows application 1332 to run on any operating system that includes a runtime environment. Similarly, operating system 1330 may support containers, and application 1332 may take the form of a container, which is a lightweight, stand-alone, executable software package that includes, for example, code for the application, a runtime, system tools, system libraries, and settings.

[0086] In addition, the computer 1302 may be enabled with a security module such as a trusted processing module ("TPM"). For example, in the case of a TPM, the boot component is hashed in the next boot component, and the result is waited for to match the security value before loading the next boot component. This process may occur at any layer in the code execution stack of the computer 1302, such as applied to the application execution level or the operating system ("OS") kernel level, thereby achieving security at any code execution level.

[0087] A user may enter commands and information into the computer 1302 through one or more wired / wireless input devices, such as a keyboard 1338, a touch screen 1340, and a pointing device such as a mouse 1342. Other input devices (not shown) may include a microphone, an infrared ("IR") remote control, a radio frequency ("RF") remote control or other remote control, a joystick, a virtual reality controller and / or a virtual reality headset, a game pad, a stylus, an image input device (e.g., a camera), a gesture sensor input device, a visual movement sensor input device, an emotion or facial detection device, a biometric input device (e.g., a fingerprint or iris scanner), and the like. These and other input devices are typically connected to the processing unit 1304 through an input device interface 1344, which may be coupled to the system bus 1308, but these and other input devices may be connected through other interfaces, such as a parallel port, an IEEE 1394 serial port, a game port, a USB port, an IR port, a fingerprint or iris scanner, or the like. Interfaces, etc.

[0088] A monitor 1346 or other type of display device may also be connected to the system bus 1308 via an interface, such as a video adapter 1348. In addition to the monitor 1346, computers typically include other peripheral output devices (not shown), such as speakers, printers, combinations thereof, and the like.

[0089] The computer 1302 can operate in a networked environment using logical connections to one or more remote computers, such as remote computer 1350, via wired and / or wireless communications. The remote computer 1350 can be a workstation, server computer, router, personal computer, portable computer, microprocessor-based entertainment device, peer device, or other public network node, and typically includes many or all of the elements described with respect to the computer 1302, but only a memory / storage device 1352 is shown for simplicity. The depicted logical connections include wired / wireless connections to a local area network ("LAN") 1354 and / or a larger network, such as a wide area network ("WAN") 1356. Such LAN and WAN networking environments are common in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to a global communication network, such as the Internet.

[0090] When used in a LAN networking environment, the computer 1302 can be connected to the local network 1354 through a wired and / or wireless communication network interface or adapter 1358. The adapter 1358 can facilitate wired or wireless communication with the LAN 1354, which can also include a wireless access point (AP) disposed thereon for communicating with the adapter 1358 in a wireless mode.

[0091] When used in a WAN networking environment, the computer 1302 may include a modem 1360 or may be connected to a communication server on the WAN 1356 via other means for establishing communications over the WAN 1356, such as through the Internet. The modem 1360, which may be internal or external and a wired or wireless device, may be connected to the system bus 1308 via the input device interface 1344. In a networked environment, program modules shown relative to the computer 1302, or portions thereof, may be stored in the remote memory / storage device 1352. It should be appreciated that the network connections shown are examples and other means of establishing a communications link between the computers may be used.

[0092] When used in a LAN or WAN networking environment, the computer 1302 may access a cloud storage system or other network-based storage system in addition to or in lieu of the external storage device 1316 described above. Generally, the connection between the computer 1302 and the cloud storage system may be established through the LAN 1354 or WAN 1356, for example, through an adapter 1358 or a modem 1360, respectively. When the computer 1302 is connected to the associated cloud storage system, the external storage interface 1326 may manage the storage provided by the cloud storage system with the assistance of the adapter 1358 and / or the modem 1360, as with other types of external storage devices. For example, the external storage interface 1326 may be configured to provide access to cloud storage sources as if those sources were physically connected to the computer 1302.

[0093] The computer 1302 may be operable to communicate with any wireless device or entity that is operatively configured to communicate wirelessly, such as a printer, scanner, desktop and / or portable computer, portable data assistant, communication satellite, any piece of equipment or location associated with a wirelessly detectable tag (e.g., a self-service machine, a newsstand, a store shelf, etc.). This may include Wireless Fidelity ("Wi-Fi") and Wireless technology. Therefore, the communication can be a predefined structure like a conventional network, or just an ad hoc communication between at least two devices.

[0094] What has been described above only includes examples of systems, computer program products, and computer-implemented methods. Of course, it is impossible to describe every conceivable combination of components, products, and / or computer-implemented methods for the purpose of describing the present disclosure, but those of ordinary skill in the art will recognize that many further combinations and permutations of the present disclosure are possible. In addition, with respect to the extent to which the terms "including", "having", "having", etc. are used in the specific embodiments, claims, appendices, and drawings, such terms are intended to be inclusive in a manner similar to the term "including", as interpreted when "including" is used as a transitional word in the claims. Descriptions of various embodiments have been given for illustrative purposes, but these descriptions are not intended to be exhaustive or limited to the disclosed embodiments. Without departing from the scope and spirit of the described embodiments, many modifications and variations will be apparent to those of ordinary skill in the art. The terms used herein are selected to best explain the principles of the embodiments, to be superior to the practical application or technical improvements of the technologies found on the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A system for a data augmentation process, the system comprising: a memory storing computer executable components; as well as a processor operatively coupled to the memory, the processor executing the computer executable components stored in the memory, wherein the computer executable components include: a data augmentation component that performs a random one-dimensional augmentation algorithm to augment a data set used to train a machine learning model via a plurality of augmentation operations, wherein the random one-dimensional augmentation algorithm employs a global augmentation parameter that defines: a distortion magnitude associated with the plurality of augmentation operations, and a number of augmentation operations included in the plurality of augmentation operations, a parameter component that generates a parameter definition for a pool of possible augmentation operations, wherein the parameter definition aligns the possible augmentation operations with the global augmentation parameter such that an amount of augmentation is directly related to the global augmentation parameter, Wherein, the data set includes images.

2. The system of claim 1 , wherein the data augmentation component randomly selects the plurality of augmentation operations from the pool of possible augmentation operations with uniform probability, and wherein the distortion magnitude controls an amount of augmentation associated with an augmentation operation from the plurality of augmentation operations.

3. The system according to claim 2, further comprising: A global parameter component generates a one-dimensional search space based on the global enhancement parameter and the selected plurality of enhancement operations.

4. The system of claim 3, wherein the parameter component controls the density of the one-dimensional search space by introducing a random uniform distribution into the parameter definition.

5. The system according to claim 4, further comprising: A search component executes a search algorithm on the one-dimensional search space based on the global enhancement parameter to perform automatic search space reduction.

6. A system according to claim 5, wherein the search algorithm performs the automatic search space reduction based on a unimodal function of a performance metric characterizing the machine learning model.

7. A computer-implemented method for a data augmentation process, the method comprising: performing, by a system operably coupled to a processor, a random one-dimensional augmentation algorithm to augment a data set used to train a machine learning model via a plurality of augmentation operations, wherein the random one-dimensional augmentation algorithm employs a global augmentation parameter that defines: a distortion magnitude associated with the plurality of augmentation operations, and a number of augmentation operations included in the plurality of augmentation operations; as well as generating, by the system, a parameter definition for a pool of possible augmentation operations, wherein the parameter definition aligns the possible augmentation operations with the global augmentation parameter such that an amount of augmentation is directly related to the global augmentation parameter, Wherein, the data set includes images.

8. The computer-implemented method for a data augmentation process of claim 7, wherein the random one-dimensional augmentation algorithm randomly selects the plurality of augmentation operations from the pool of possible augmentation operations with uniform probability, and wherein the distortion amplitude controls an amount of augmentation associated with an augmentation operation from the plurality of augmentation operations.

9. The computer-implemented method for a data enhancement process of claim 8, further comprising: A one-dimensional search space is generated by the system based on the global enhancement parameter and the selected plurality of enhancement operations.

10. A computer-implemented method for a data augmentation process according to claim 9, wherein said generating said parameter definition controls the density of said one-dimensional search space by introducing a random uniform distribution into said parameter definition.

11. The computer-implemented method for a data augmentation process of claim 10, further comprising: A search algorithm is executed by the system on the one-dimensional search space based on the global enhancement parameter to perform automatic search space reduction.

12. A computer-implemented method for a data augmentation process according to claim 11, wherein the search algorithm performs the automatic search space reduction based on a unimodal function that characterizes a performance metric of the machine learning model.

13. A computer program product for a data enhancement process, the computer program product comprising a computer readable storage medium having program instructions embodied therein, the program instructions being executable by a processor to cause the processor to: executing, by the processor, a random one-dimensional augmentation algorithm to augment a data set used to train a machine learning model via a plurality of augmentation operations, wherein the random one-dimensional augmentation algorithm employs a global augmentation parameter that defines: a distortion magnitude associated with the plurality of augmentation operations, and a number of augmentation operations included in the plurality of augmentation operations; and generating, by the processor, a parameter definition for a pool of possible augmentation operations, wherein the parameter definition aligns the possible augmentation operations with the global augmentation parameter such that an amount of augmentation is directly related to the global augmentation parameter, in, The dataset includes images.

14. The computer program product of claim 13, wherein the random one-dimensional enhancement algorithm randomly selects the plurality of enhancement operations from the pool of possible enhancement operations with uniform probability, and wherein the distortion magnitude controls an amount of enhancement associated with an enhancement operation from the plurality of enhancement operations.

15. The computer program product of claim 14, wherein the program instructions further cause the processor to: A one-dimensional search space is generated by the processor based on the global enhancement parameter and the selected plurality of enhancement operations.

16. The computer program product of claim 15, wherein the density of the one-dimensional search space is controlled by introducing a random uniform distribution into the parameter definition.

17. The computer program product of claim 16, wherein the program instructions further cause the processor to: A search algorithm is executed by the processor on the one-dimensional search space based on the global enhancement parameter to perform automatic search space reduction, wherein the search algorithm performs the automatic search space reduction based on a unimodal function characterizing a performance metric of the machine learning model.

Images