AI Learning Module Using Generative Adversarial Networks for Synthetic Data Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training artificial intelligence models using synthetic data face challenges such as lower learning accuracy, information loss due to stride size, and increased storage requirements, making them less effective and more resource-intensive compared to using real data.
Innovation Solution
An electronic device is equipped with a learning module that includes a first generator for creating input vectors, a second generator for producing synthetic data, and a first learning model for analyzing this synthetic data. The device learns parameters from both generators based on output data, minimizes loss values, and optionally replaces convolutional layers with swing convolution layers to prevent information loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If virtual data is used for training AI networks, then learning operation is enabled when real data is unavailable due to security or cost issues, but learning accuracy deteriorates compared to using real data
Solution Approach 1:
The patent introduces a discriminator network as an intermediary component that evaluates the authenticity of synthetic data and provides feedback to the generator. This mediator mechanism allows the system to improve synthetic data quality iteratively, bridging the gap between virtual data availability and learning accuracy requirements
Solution Approach 2:
The patent implements a feedback loop where the discriminator network assesses synthetic data quality and returns signals to the generator for refinement. This feedback mechanism enables continuous improvement of synthetic data, allowing the system to achieve acceptable learning accuracy despite using virtual data
2Productivity
If learning operation is performed using virtual data, then training can proceed without real data, but information loss occurs depending on stride size and learning performance deteriorates
Solution Approach 1:
The patent dynamically adjusts the stride size parameter of convolutional layers during training based on the quality of synthetic data and learning progress. By changing this critical parameter, the system optimizes information retention while maintaining training efficiency with virtual data
Solution Approach 2:
The patent introduces dynamic adjustment mechanisms that allow the generator and discriminator networks to adapt their parameters in real-time during training. This dynamic behavior enables the system to optimize information flow and reduce information loss caused by fixed stride sizes
3Object-affected harmful factors
If using virtual data for training, then security and cost issues are avoided, but storage size of pre-learning model increases making it difficult to use on terminal devices
Solution Approach 1:
The patent extracts and removes redundant information from the training process by using a more efficient generator architecture that produces high-quality synthetic data with less computational overhead. This extraction of essential information allows for smaller model sizes while maintaining the security and cost benefits of virtual data
Data Source
AI summary
An electronic device is provided. The electronic device includes memory storing a learning module comprising a first generator that generates an input vector, a second generator that generates synthetic data, a first learning model that analyzes the synthetic data and one or more computer programs, and one or more processors communicatively coupled to the memory, wherein the one or more computer programs include computer-executable instructions that, when executed by the one or more processors individually or collectively, cause the electronic device to obtain the input vector from the first generator, obtain the synthetic data corresponding to the input vector through inputting the input vector into the second generator, obtain output data by analyzing the synthetic data by inputting the synthetic data into the first learning model, and learn, at least one parameter included in the first generator and at least one parameter included in the second generator on the basis of the output data.