Autonomous learning training method for AI digital asset content

Abstract

The invention relates to the technical field of artificial intelligence model training, and discloses an autonomous learning training method for AI digital asset content, and the method comprises the steps: firstly obtaining original data, carrying out the preprocessing of the original data, obtaining a data feature matrix, generating initial feature representation through an adaptive multi-layer perception network, and carrying out the deep reinforcement learning optimization, and generating synthetic data through the generative adversarial network, and adjusting network parameters through a comparative learning algorithm to obtain final digital asset content for training an AI model. The method effectively solves the problem of existing AI digital asset content processing, can improve the data quality and utilization efficiency, expand the data volume, enhance the model generalization ability, optimize the model training process, improve the training precision, meet the real-time and safety requirements, reduce the labor cost, improve the training autonomy, and provide powerful support for AI technology development.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence model training technology, and specifically to an autonomous learning training method for AI digital asset content. Background Art

[0002] As artificial intelligence (AI) is booming, the training of AI models is highly dependent on a large amount of high-quality data. As a key data source, the processing and utilization of digital asset content is crucial to the development of AI. However, the current AI digital asset content processing faces many difficulties, which seriously restricts the further breakthrough and application expansion of AI technology.

[0003] Uneven data quality is a prominent problem. In reality, digital assets come from a wide range of sources and in various formats, such as pictures, text, audio, and video. Taking image digital assets as an example, they may be shot at different resolutions, different shooting equipment, and different lighting conditions, resulting in huge differences in image data clarity, color accuracy, and noise levels; text digital assets may have spelling errors, grammatical errors, semantic ambiguity, and inconsistent encoding formats; audio digital assets may be affected by recording environment noise interference, inconsistent sampling rates, etc. These quality issues make it difficult to use raw data directly for efficient AI model training. If low-quality data is used directly, it will lead to poor model training results, low accuracy, and poor generalization capabilities.

[0004] Insufficient data is also a key factor limiting the performance improvement of AI models. In many specific fields, such as medical image analysis and rare species identification, high-quality digital asset data is extremely scarce. The acquisition of medical imaging data not only requires professional equipment and complex processes, but also involves many restrictions such as patient privacy, which leads to slow data accumulation; and for image or audio data of rare species, it is extremely difficult to collect due to the small number of species themselves and the special living environment. The limited amount of data makes it difficult for AI models to learn comprehensive and accurate feature patterns. When faced with new data, the model lacks sufficient generalization ability and cannot accurately perform tasks such as classification, prediction or generation.

[0005] Traditional AI model training methods have dual bottlenecks in efficiency and accuracy when processing digital asset content. On the one hand, traditional data preprocessing methods are often based on fixed rules or simple statistical methods, and it is difficult to perform adaptive processing based on the complex characteristics of the data. For example, when normalizing image data, the commonly used fixed range normalization method may not be able to fully retain the key details in the image, resulting in the loss of important features in subsequent model training; on the other hand, in the model training process, traditional supervised learning and unsupervised learning methods rely on a large number of manual annotations or preset model structures, which not only consumes a lot of manpower and time costs, but also easily falls into local optimal solutions when processing complex and diverse digital asset content, and cannot fully tap the potential value in the data. With the continuous expansion and deepening of AI application scenarios, the real-time and security requirements for digital asset content processing are becoming higher and higher. In scenarios such as real-time video surveillance and real-time warning of financial transaction risks, a large amount of digital asset data needs to be quickly processed and analyzed, and traditional processing methods are difficult to meet this real-time requirement; at the same time, the digital asset content contains a large amount of sensitive information, such as personal identity information, commercial secrets, etc. How to ensure the security and privacy of data during AI training and prevent data leakage and abuse has become an important issue that needs to be solved urgently. Summary of the invention

[0006] The purpose of the present invention is to provide an autonomous learning training method for AI digital asset content to solve the problems raised in the above background technology.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solution: an autonomous learning training method for AI digital asset content, the method comprising: Acquire original data of the digital asset content, and preprocess the original data to obtain a preprocessed data feature matrix; According to the preprocessed data feature matrix, generating an initial feature representation based on an adaptive multi-layer perception network; Using the initial feature representation, optimizing the features through a deep reinforcement learning framework to obtain an optimized feature representation; Generate synthetic data of digital asset content based on a generative adversarial network according to the optimized feature representation; Using the synthetic data to compare with the original data, and adjusting the parameters of the generative adversarial network through a contrastive learning algorithm; Generate the final digital asset content based on the adjusted generative adversarial network; The final digital asset content is used to perform model training to obtain a trained AI model.

[0008] Preferably, generating an initial feature representation based on an adaptive multi-layer perceptron network according to the preprocessed data feature matrix comprises: Normalizing the preprocessed data feature matrix to obtain a normalized data feature matrix; Using the normalized data feature matrix, a nonlinear transformation is performed through the hidden layer of the adaptive multi-layer perception network. The calculation formula is: in, is the hidden layer output, is the activation function, is the hidden layer weight matrix, is the hidden layer bias vector, is the normalized data feature matrix; The hidden layer output is linearly transformed through the output layer to obtain the initial feature representation. The calculation formula is: in, is the initial feature representation, is the output layer weight matrix, is the output layer bias vector.

[0009] Preferably, the using the initial feature representation to optimize the features through a deep reinforcement learning framework to obtain the optimized feature representation includes: Define the state space of the deep reinforcement learning framework as the initial feature representation, and the action space as the set of adjustment operations on the features; Construct a reward function that is related to the diversity and effectiveness of features; The proximal policy optimization algorithm is used to perform iterative optimization in the deep reinforcement learning framework, update the parameters of the policy network and the value network, and obtain the optimized policy network; The optimized policy network is used to operate on the initial feature representation to obtain the optimized feature representation.

[0010] Preferably, generating synthetic data of digital asset content based on a generative adversarial network according to the optimized feature representation includes: The optimized feature representation is input into the generator of the generative adversarial network. The generator consists of multiple layers of transposed convolutional layers. The calculation formula is: in, For the generator, is the optimized feature representation, For the Layer transposed convolutional layer; After the generator generates synthetic data, the discriminator distinguishes the synthetic data from the real data. The discriminator consists of multiple convolutional layers, and the calculation formula is: in, is the discriminator, For input data, For the Layer convolutional layer; Through adversarial training, the parameters of the generator and discriminator are continuously adjusted to enable the generator to generate more realistic synthetic data.

[0011] Preferably, comparing the synthetic data with the original data and adjusting the parameters of the generative adversarial network by a contrastive learning algorithm comprises: Calculate the cosine similarity between the synthetic data and the original data, the calculation formula is: Among them, sim is the cosine similarity, and The synthetic data and the original data are The values ​​in the dimensions, is the data dimension; the contrast loss function is constructed based on the cosine similarity, and the calculation formula is: in, is the contrast loss function, is the temperature parameter, is the number of data samples, is the cosine similarity of other samples; The contrastive loss function is used to adjust the parameters of the generator and discriminator of the generative adversarial network.

[0012] Preferably, generating the final digital asset content according to the adjusted generative adversarial network includes: The optimized feature representation is input into the generator of the adjusted generative adversarial network again; The generator outputs the final digital asset content, and the calculation formula is: in, For the final digital asset content, is the adjusted generator, is the optimized feature representation.

[0013] Preferably, the using the final digital asset content to perform model training to obtain a trained AI model includes: Select convolutional neural network as the architecture of the target AI model; The final digital asset content is used as training data and input into the target AI model for training. The parameters of the target AI model are updated using the stochastic gradient descent algorithm. The calculation formula is: in, is the updated parameter, is the current parameter, is the learning rate, is the gradient of the loss function with respect to the current parameters; after multiple rounds of training, a trained AI model is obtained.

[0014] Preferably, the method further includes the step of performing a quality assessment on the final digital asset content after it is generated: calculating the information entropy of the final digital asset content to assess the uncertainty and richness of the data; calculating the structural similarity index between the final digital asset content and the original data to assess the degree of structural similarity between the two; and judging whether the quality of the final digital asset content meets the requirements based on the information entropy and the structural similarity index.

[0015] Preferably, if the quality of the final digitized asset content does not meet the requirements, the process returns to the step of preprocessing the original data and performs subsequent operations again.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention effectively solves the problem of uneven data quality by preprocessing the original data of the digital asset content and converting it into a unified and standardized data feature matrix. In the processing of image digital assets, the resolution and color mode are unified and noise interference is removed, so that subsequent model training can focus on key features, greatly improving the availability of data. At the same time, based on the adaptive multi-layer perception network to generate the initial feature representation, key features can be adaptively extracted according to the complex characteristics of the data. Compared with the traditional fixed rule feature extraction method, more effective information is retained, which further improves the data quality and enhances the model's understanding and utilization efficiency of the data.

[0017] The deep reinforcement learning framework is used for feature optimization, and the synthetic data of digital asset content is generated in combination with the generative adversarial network, which successfully expands the amount of data. In areas where data is scarce, such as medical image analysis and rare species research, the diversity of training samples is increased by synthetic data. The generator and discriminator in the generative adversarial network are trained against each other, so that the generated synthetic data is highly similar to the real data in terms of features and distribution. This not only enriches the material for model training, but also allows the model to learn a wider range of feature patterns, significantly enhances the generalization ability of the model, and improves the performance of the model on new data. The contrastive learning algorithm used in the invention adjusts the parameters of the generative adversarial network, which can optimize the generation process according to the similarity between the synthetic data and the original data, so that the generated data is closer to the real data. During model training, the final digital asset content is used as training data, and the target AI model parameters are updated in combination with algorithms such as stochastic gradient descent. This method avoids the problem that traditional training methods are prone to falling into local optimal solutions. Taking convolutional neural networks as an example, in image classification tasks, the accuracy of the model trained by the method of the present invention is significantly improved compared with traditional methods, and can more accurately identify the object categories in the image and reduce misclassification.

[0018] In terms of real-time performance, the autonomous learning training method of the present invention has been optimized and has a faster processing speed. In real-time video surveillance scenarios, video data can be quickly preprocessed, feature extracted, and model trained, abnormal behavior can be discovered in time, and early warning can be issued, meeting the requirements of real-time processing in practical applications. In terms of security, the entire training process has a more complete data processing and protection mechanism. When processing digital asset content containing sensitive information, the risks caused by data leakage and abuse are avoided, the security and privacy of data are guaranteed, and strong support is provided for the application of AI in fields such as finance and medical care that have extremely high data security requirements. Unlike traditional training methods that rely on a large number of manual annotations and preset model structures, the autonomous learning training method of the present invention can automatically learn features and rules from data. In the processing of text digital assets, there is no need to manually annotate text sentiment tendencies, semantic categories, and other information one by one. The model can autonomously learn these features, reducing a lot of manpower investment. At the same time, the entire training process has higher autonomy, and can adaptively adjust the training strategy according to data characteristics and training effects, improve training efficiency and quality, reduce technical dependence on professionals, and make AI model training more convenient and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a working principle diagram of the autonomous learning training method of the present invention; Figure 2 A diagram of implementation steps for utilizing the initial feature representation to optimize the features through a deep reinforcement learning framework to obtain an optimized feature representation; Figure 3A diagram of the implementation steps for generating synthetic data based on a generative adversarial network; Figure 4 A diagram showing the implementation steps for adjusting the parameters of a generative adversarial network using a contrastive learning algorithm. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] See also Figure 1-4 The present invention provides a technical solution: an autonomous learning training method for AI digital asset content, the method comprising: The original data of the digital asset content is obtained, and the original data is preprocessed to obtain the preprocessed data feature matrix. For example, in an image recognition project, the original data is a large number of pictures, and these pictures are preprocessed by denoising, normalizing the size, etc., and each picture is converted into a feature matrix of fixed dimensions for subsequent processing.

[0022] According to the preprocessed data feature matrix, an initial feature representation is generated based on an adaptive multi-layer perception network. The network is used to perform nonlinear and linear transformations on the preprocessed data to extract the initial features of the data.

[0023] The initial feature representation is used to optimize the features through a deep reinforcement learning framework to obtain an optimized feature representation. The features are optimized using a proximal strategy optimization algorithm by defining a suitable state space, action space, and reward function.

[0024] According to the optimized feature representation, synthetic data of digital asset content is generated based on a generative adversarial network. The generator and the discriminator are trained against each other to generate more realistic synthetic data.

[0025] The synthetic data is compared with the original data, and the parameters of the generative adversarial network are adjusted through a contrastive learning algorithm. The cosine similarity between the two is calculated and a contrastive loss function is constructed to optimize the generative adversarial network.

[0026] The final digital asset content is generated according to the adjusted generative adversarial network. The optimized feature representation is input into the generator of the adjusted generative adversarial network again to obtain the final digital asset content.

[0027] The final digital asset content is used for model training to obtain a trained AI model. A suitable neural network architecture is selected, the final digital asset content is used for training, and the model parameters are updated using a stochastic gradient descent algorithm.

[0028] The present invention will be further described below in conjunction with Examples 1 to 5: Embodiment 1: This embodiment describes in detail how to process the preprocessed data based on an adaptive multi-layer perception network to generate an initial feature representation, which is the basis for subsequent feature optimization and data generation, ensuring that valuable initial feature information is extracted from the original data.

[0029] In practical applications, taking a speech recognition project as an example, assuming that the original data of the digital asset content obtained is a large number of speech audio files. First, these audio files are preprocessed, and the audio signals are sampled, filtered, and converted into preprocessed data feature matrices.

[0030] The preprocessed data feature matrix is ​​normalized to obtain a normalized data feature matrix. Normalization is to make the data in the same scale range to facilitate model learning. For example, the maximum-minimum normalization method is used to map the data to the [0,1] interval. Assume that a certain eigenvalue in the original data feature matrix is , the maximum value of the feature is max, and the minimum value is min, then the normalized feature value The calculation formula is: Using the normalized data feature matrix, the nonlinear transformation is performed through the hidden layer of the adaptive multilayer perceptron. The adaptive multilayer perceptron has the ability to learn and adjust its own parameters to adapt to the data characteristics. Taking an adaptive multilayer perceptron with one hidden layer as an example, the hidden layer output The calculation formula is: Among them, assuming that the activation function Select the ReLU function, that is Hidden layer weight matrix and the hidden layer bias vector Continuously adjust and optimize during the training process. Assume that the normalized data feature matrix is a size of The matrix of represents the number of samples, Represents the feature dimension. Hidden layer weight matrix The size is , is the number of hidden layer neurons, the hidden layer bias vector The size is After calculation, the hidden layer output is obtained , whose size is .

[0031] The hidden layer output is linearly transformed through the output layer to obtain the initial feature representation. The calculation formula is: Output layer weight matrix The size is , is the dimension of the initial feature representation of the output, the output layer bias vector The size is Through this linear transformation, the output of the hidden layer is transformed into the final initial feature representation , whose size is In speech recognition projects, the initial feature representation obtained can include multiple feature information such as speech frequency, duration, and intonation, providing an important data basis for subsequent feature optimization and model training.

[0032] Embodiment 2: This embodiment explains in detail how to use the deep reinforcement learning framework to optimize the initial feature representation. By defining a suitable state space, action space, and reward function, and adopting a proximal strategy optimization algorithm, the optimized feature representation is made more diverse and effective, thereby improving the performance of the model.

[0033] In an actual scenario of image classification, the deep reinforcement learning framework is used for feature optimization. First, the state space of the deep reinforcement learning framework is defined as the initial feature representation. For example, the initial feature representation obtained in Example 1 is a Each row represents the initial features of an image sample, and each column represents a different feature dimension. This matrix is ​​used as the state of the deep reinforcement learning framework, and the agent makes decisions in this state space.

[0034] The action space is a set of adjustment operations on features. For image features, actions can include scaling, translating, adding noise, and other operations on certain feature dimensions. For example, an action is defined as multiplying the values ​​of certain feature dimensions by a to achieve the scaling operation of the feature; or add a random number between A random number between and is used to implement the translation operation.

[0035] Construct a reward function that is related to the diversity and effectiveness of features. In image classification tasks, the diversity of features can be measured by calculating the correlation between features. The lower the correlation, the higher the diversity. The effectiveness of features can be measured by the accuracy of image classification using the current features. Assume that the features in the current state are represented as , define the reward function for: in, represents the correlation coefficient between features, Indicates the use of features The accuracy of image classification, and is a weight coefficient used to balance the importance of diversity and effectiveness in the reward function. For example, setting .

[0036] The proximal policy optimization algorithm is used to perform iterative optimization in the deep reinforcement learning framework, update the parameters of the policy network and the value network, and obtain the optimized policy network. The proximal policy optimization algorithm makes the training more stable and efficient by limiting the amplitude of the policy update. During the training process, the agent selects actions from the policy network according to the current state, and obtains a new state and reward after executing the action. Based on this data, the parameters of the policy network and the value network are updated. Assume that the parameters of the policy network are , the parameters of the value network are , through multiple iterations, continuous adjustment and , which enables the policy network to select better actions.

[0037] The optimized policy network is used to operate on the initial feature representation to obtain the optimized feature representation. In the image classification scenario, the optimized policy network selects a series of actions based on the initial feature representation, such as scaling and translating certain feature dimensions. After these operations, the optimized feature representation is obtained, which has improved diversity and effectiveness, and is more conducive to subsequent image classification tasks.

[0038] Embodiment 3: This embodiment specifically describes how to generate synthetic data of digital asset content based on a generative adversarial network using optimized feature representation, continuously improve the quality of synthetic data through adversarial training of the generator and the discriminator, and provide rich data for subsequent comparative learning and model training.

[0039] Take a project to generate handwritten digital images as an example, and generate synthetic data of digital asset content based on the generative adversarial network. The optimized feature representation is input into the generator of the generative adversarial network. The generator consists of multiple layers of transposed convolutional layers, and the calculation formula is: Assume that the generator has 3 layers of transposed convolutional layers, the first layer of transposed convolutional layer The input is the optimized feature representation , whose size is (Assuming that the optimized feature representation is compressed to such a dimension), after the first transposed convolution layer, the output size becomes , the convolution kernel size is , the step size is 1; the second transposed convolution layer The input is the output of the first layer. After passing through this layer, the output size becomes , the convolution kernel size is , the step size is 2; the third transposed convolution layer The input is the output of the second layer. After passing through this layer, the output size becomes , what we get is the generated handwritten digital image (grayscale image).

[0040] After the generator generates synthetic data, the discriminator distinguishes the synthetic data from the real data. The discriminator consists of multiple convolutional layers, and the calculation formula is: Assume that the discriminator has 4 convolutional layers, the first convolutional layer The input is of size The image data (can be synthetic data or real data) is , the step size is 2, and the output size becomes The second convolutional layer The input is the output of the first layer, and the convolution kernel size is , the step size is 2, and the output size becomes The third convolutional layer The input is the output of the second layer, and the convolution kernel size is , the step size is 2, and the output size becomes ; The fourth convolutional layer The input is the output of the third layer, and after passing through this layer, a scalar is output to determine whether the input image is real data or synthetic data.

[0041] Through adversarial training, the parameters of the generator and the discriminator are continuously adjusted to make the generator generate more realistic synthetic data. During the training process, the generator tries to generate synthetic data that can deceive the discriminator, while the discriminator tries to accurately distinguish between real data and synthetic data. Assume that the loss function of the generator is , the loss function of the discriminator is ,The adversarial training algorithm used is the minimum-maximum game algorithm. The goal of the generator is to minimize ,Right now in It is real data. is random noise (replaced here by the optimized feature representation), is the distribution of the real data, is the distribution of noise. The goal of the discriminator is to maximize ,Right now Through continuous iterative training, the handwritten digit images generated by the generator become more and more realistic and can better simulate realistic handwritten digit data.

[0042] Embodiment 4: This embodiment aims to construct a loss function based on the comparison of synthetic data and original data through a contrastive learning algorithm, and optimize and adjust the parameters of the generative adversarial network to improve the quality of the generated data, make it more similar to the original data at the feature level, and enhance the effectiveness and authenticity of the data generated by the generative adversarial network.

[0043] Taking the facial image generation task as an example, assuming that we have generated a batch of synthetic facial image data through a generative adversarial network, we now need to use a contrastive learning algorithm to adjust the parameters of the generative adversarial network.

[0044] Calculate the cosine similarity between the synthesized data and the original data. Convert both the original facial image and the synthesized facial image into feature vectors, for example, using a pre-trained convolutional neural network (such as VGG16) to extract the features of the image. Assume that the original image feature vector is , the synthetic image feature vector is ,here is the dimension of the feature vector, assuming According to the cosine similarity formula For example, for a pair of original images and synthesized images, the calculated cosine similarity is is 0.6.

[0045] Then, we construct a contrast loss function based on cosine similarity. Setting temperature parameters , in a batch containing 100 image samples (including original images and synthesized images), Take 20 (i.e. select 20 sample pairs for comparison calculation). The calculation formula of the comparison loss function is: in is the cosine similarity of other sample pairs. In the actual calculation, we traverse these 20 sample pairs, calculate their cosine similarities respectively, and then substitute them into the formula to calculate the contrast loss function value. Assume that the calculated contrast loss function value is 0.8. Finally, the contrast loss function is used to adjust the parameters of the generator and discriminator of the generative adversarial network. Through the back-propagation algorithm, the gradient of the contrast loss function is passed to the generator and discriminator. Assume that the parameters of the generator are , the parameters of the discriminator are , use Adam optimizer to update parameters. The learning rate is set to 0.0001. For the generator, the parameter update formula is in is the learning rate, is the training round; for the discriminator, the parameter update formula is After multiple rounds of training, the parameters of the generator and discriminator are continuously adjusted so that the generated facial images are closer to the original facial images in terms of details and distribution of facial features.

[0046] Embodiment 5: This embodiment mainly demonstrates how to use the final generated digital asset content to train the AI ​​model and obtain training results that meet actual needs. At the same time, the final digital asset content is evaluated through specific quality assessment indicators, and the decision on whether to re-process the data and model training process is made based on the evaluation results to ensure the reliability and practicality of the entire technical solution.

[0047] In the target detection project under the autonomous driving scenario, the final digital asset content is used for model training and quality assessment.

[0048] Select convolutional neural network (CNN) as the architecture of the target AI model. Here, the classic YOLO (You Only Look Once) series of networks, such as YOLOv5, are selected. The final digital asset content is used as training data. These data are synthetic images containing vehicles, pedestrians, traffic signs and other targets generated after a series of previous processing. During the training process, these synthetic images and their corresponding annotation information (target category, location, etc.) are input into the YOLOv5 model.

[0049] The stochastic gradient descent algorithm is used to update the parameters of the target Al model. The formula is: in is the current parameter, is the updated parameter, is the learning rate, set , is the gradient of the loss function with respect to the current parameters. The loss function combines the classification loss (such as cross entropy loss) and the positioning loss (such as mean square error loss) in the target detection task. During the training process, the model continuously adjusts the parameters according to the feedback of the loss function. After 500 rounds of training, the model gradually converges and the detection accuracy of various targets continues to improve.

[0050] After the final digital asset content is generated, its quality is evaluated. The information entropy of the final digital asset content is calculated to evaluate the uncertainty and richness of the data. For the synthesized autonomous driving scene image, the pixel value of the image is regarded as a probability distribution, and the information entropy formula is used to calculate the probability distribution of the pixel value. Calculate information entropy, where Is the pixel value The probability of . Suppose the information entropy of a synthetic image is calculated as The higher the information entropy, the more uniform the distribution of pixel values ​​in the image and the richer the data.

[0051] Calculate the structural similarity index (SSIM) of the final digital asset content and the original data to evaluate the structural similarity between the two. SSIM measures similarity by comparing the brightness, contrast and structural information of the image. The value range is between [-1,1]. The closer the value is to 1, the more similar the structure is. Use a special SSIM calculation function to calculate the synthetic image and the original image. Assume that the obtained SSIM value is .

[0052] According to the information entropy and structural similarity index, it is judged whether the quality of the final digital asset content meets the requirements. The threshold of information entropy is set to bits, the SSIM threshold is If the information entropy of a synthetic image and , then the quality of the synthesized image meets the requirements; if not, for example, the information entropy of an image is 6.5 bits and the SSIM value is 0.75, then return to the step of preprocessing the original data and perform subsequent operations again to improve the quality of the synthesized data, thereby improving the performance of the target detection model.

[0053] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0054] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An autonomous learning training method for AI digital asset content, characterized in that: The method comprises: Acquire original data of the digital asset content, and preprocess the original data to obtain a preprocessed data feature matrix; According to the preprocessed data feature matrix, generating an initial feature representation based on an adaptive multi-layer perception network; Using the initial feature representation, optimizing the features through a deep reinforcement learning framework to obtain an optimized feature representation; Generate synthetic data of digital asset content based on a generative adversarial network according to the optimized feature representation; Comparing the synthetic data with the original data, and adjusting the parameters of the generative adversarial network through a contrastive learning algorithm; Generate the final digital asset content based on the adjusted generative adversarial network; The final digital asset content is used to perform model training to obtain a trained AI model.

2. The autonomous learning training method for AI digital asset content according to claim 1, characterized in that: Generating an initial feature representation based on an adaptive multi-layer perception network according to the preprocessed data feature matrix includes: Normalizing the preprocessed data feature matrix to obtain a normalized data feature matrix; Using the normalized data feature matrix, a nonlinear transformation is performed through the hidden layer of the adaptive multi-layer perception network. The calculation formula is: in, is the hidden layer output, is the activation function, is the hidden layer weight matrix, is the hidden layer bias vector, is the normalized data feature matrix; The hidden layer output is linearly transformed through the output layer to obtain the initial feature representation. The calculation formula is: in, is the initial feature representation, is the output layer weight matrix, is the output layer bias vector.

3. The autonomous learning training method for AI digital asset content according to claim 2, characterized in that: The utilizing the initial feature representation and optimizing the feature through a deep reinforcement learning framework to obtain an optimized feature representation includes: Define the state space of the deep reinforcement learning framework as the initial feature representation, and the action space as the set of adjustment operations on the features; Construct a reward function that is related to the diversity and effectiveness of features; The proximal policy optimization algorithm is used to perform iterative optimization in the deep reinforcement learning framework, update the parameters of the policy network and the value network, and obtain the optimized policy network; The optimized policy network is used to operate on the initial feature representation to obtain the optimized feature representation.

4. The autonomous learning training method for AI digital asset content according to claim 3, characterized in that: The step of generating synthetic data of digital asset content based on a generative adversarial network according to the optimized feature representation includes: The optimized feature representation is input into the generator of the generative adversarial network. The generator consists of multiple layers of transposed convolutional layers. The calculation formula is: in, For the generator, is the optimized feature representation, For the Layer transposed convolutional layer; After the generator generates synthetic data, the discriminator distinguishes the synthetic data from the real data. The discriminator consists of multiple convolutional layers, and the calculation formula is: in, is the discriminator, For input data, For the Layer convolutional layer; Through adversarial training, the parameters of the generator and discriminator are continuously adjusted to enable the generator to generate more realistic synthetic data.

5. The autonomous learning training method for AI digital asset content according to claim 4, characterized in that: The step of comparing the synthetic data with the original data and adjusting the parameters of the generative adversarial network by a contrastive learning algorithm includes: Calculate the cosine similarity between the synthetic data and the original data, the calculation formula is: Among them, sim is the cosine similarity, and The synthetic data and the original data are The values ​​in the dimensions, is the data dimension; the contrast loss function is constructed based on the cosine similarity, and the calculation formula is: in, is the contrast loss function, is the temperature parameter, is the number of data samples, is the cosine similarity of other samples; The contrastive loss function is used to adjust the parameters of the generator and discriminator of the generative adversarial network.

6. The autonomous learning training method for AI digital asset content according to claim 5, characterized in that: The final digital asset content generated according to the adjusted generative adversarial network includes: The optimized feature representation is input into the generator of the adjusted generative adversarial network again; The generator outputs the final digital asset content, and the calculation formula is: in, For the final digital asset content, is the adjusted generator, is the optimized feature representation.

7. The autonomous learning training method for AI digital asset content according to claim 6, characterized in that: The method of using the final digital asset content to perform model training to obtain a trained AI model includes: Select convolutional neural network as the architecture of the target AI model; The final digital asset content is used as training data and input into the target AI model for training. The parameters of the target AI model are updated using the stochastic gradient descent algorithm. The calculation formula is: in, is the updated parameter, is the current parameter, is the learning rate, is the gradient of the loss function with respect to the current parameters; after multiple rounds of training, a trained AI model is obtained.

8. The autonomous learning training method for AI digital asset content according to claim 1, characterized in that: It also includes the steps of evaluating the quality of the final digital asset content after it is generated: calculating the information entropy of the final digital asset content and evaluating the uncertainty and richness of the data; Calculate the structural similarity index between the final digital asset content and the original data to evaluate the structural similarity between the two; Based on the information entropy and structural similarity index, determine whether the quality of the final digital asset content meets the requirements.

9. The autonomous learning training method for AI digital asset content according to claim 8, characterized in that: If the quality of the final digital asset content does not meet the requirements, return to the step of preprocessing the original data and perform subsequent operations again.

Images