Federated Learning Data Augmentation via Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Federated learning environments face accuracy degradation due to data heterogeneity among clients, and existing data sharing techniques risk exposing internal data and causing overfitting, necessitating methods that enhance data distribution using external data for improved model accuracy and training speed.
Innovation Solution
A method involving a neural network model that screens data for similarity using embedding vectors and performs backpropagation with rewards based on correlation, filtering noise and augmenting training data while preserving internal data distribution through reinforcement learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data sharing techniques are used to improve model accuracy in federated learning, then accuracy is improved, but internal data privacy is exposed and overfitting risk increases
Solution Approach 1:
The patent introduces an intermediary data augmentation mechanism that processes external public data through a neural network model to generate synthetic training data. This intermediary process allows the system to benefit from external data distributions without directly exposing or sharing sensitive internal client data, thus maintaining privacy while improving model accuracy through enhanced data diversity
Solution Approach 2:
The patent creates synthetic copies of training data by using a neural network model to generate augmented data samples that mimic the distribution characteristics of internal client data. These synthetic copies serve as proxies for actual client data, allowing the global model to learn from diverse data patterns without requiring access to or sharing of the original sensitive internal data
2Measurement precision
If external data is used to augment training data distribution, then model accuracy is improved, but data heterogeneity among clients increases
Solution Approach 1:
The patent dynamically adjusts the parameters of the data augmentation process by using reinforcement learning to optimize the neural network model's data generation. The system modifies data distribution parameters adaptively based on feedback from model performance, allowing it to balance the incorporation of external data with maintaining compatibility across heterogeneous client data distributions
Solution Approach 2:
The patent implements a dynamic data augmentation strategy where the amount and type of external data incorporated into training is adjusted based on real-time model performance feedback. The reinforcement learning component continuously adapts the data augmentation process to optimize for both accuracy improvement and compatibility with diverse client data characteristics
3Productivity
If reinforcement learning-based data augmentation is performed to improve training speed, then training efficiency is improved, but computational complexity increases
Solution Approach 1:
The patent performs data augmentation in advance during the training process rather than on-demand during inference. By pre-generating augmented training data using reinforcement learning-based data augmentation, the system prepares enhanced training datasets beforehand, which speeds up the actual model training without requiring complex computations during the training phase
Solution Approach 2:
The patent implements a self-service mechanism where the neural network model automatically generates its own augmented training data through reinforcement learning. The system uses the model's own predictions and performance feedback to guide the data augmentation process, eliminating the need for external manual data processing and reducing overall computational overhead
Data Source
AI summary
The present disclosure relates to a method for training a neural network model. First screening data is determined by inputting first data not associated with at least one client into the neural network model to calculate similarity between the first data and second data associated with the at least one client. the neural network model is trained by performing backpropagation on the neural network model with reference to a reward determined based on a correlation between the first screening data and the second data.