Machine Learning Training System with Abnormality Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning systems face instability and noise susceptibility in online learning due to rapid user behavior changes and asynchronous data updates, leading to model failures and poor user experience.
Innovation Solution
A training method that allocates data to multiple working machines, processes each data piece separately, calculates local weights and loss function values, aggregates them, and performs abnormality detection to correct or discard updates, ensuring stable model training and noise resistance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If online learning continuously updates the model with new training data, then the model accuracy is improved and user preference changes are captured, but the model becomes unstable and prone to noise influence
Solution Approach 1:
The system performs preliminary actions by collecting and storing training data before actual model training occurs. The training data collection module continuously gathers user behavior data and stores it in a database, preparing the data in advance so that when training is needed, stable and sufficient data is already available, preventing the model from being influenced by noisy real-time data
Solution Approach 2:
The system segments the training process into distinct phases: data collection, data storage, and model training. By separating these functions into different modules (training data collection module, training data storage module, model training module), the system can collect data over time without immediately training the model, thus avoiding noise while maintaining accuracy
2Adaptability or versatility
If the model is updated asynchronously with continuously flowing training data, then the model adapts to user behavior changes, but training failures occur due to uncontrollable issues
Solution Approach 1:
The system performs preliminary data collection and validation before model training. The training data collection module continuously gathers user behavior data and stores it in a database, preparing clean and validated data in advance. This preliminary preparation ensures that when model training occurs, the data is already verified and ready, preventing training failures
Solution Approach 2:
The system implements feedback mechanisms where the model training module continuously monitors training progress and model performance. When the model's predicted values deviate from actual values beyond a threshold, the system automatically triggers retraining with newly collected data, creating a stable feedback loop that adapts the model without causing training failures
3Reliability
If offline learning is used to train the model, then the model training is stable, but the model cannot capture rapid user behavior changes
Solution Approach 1:
The system maintains continuous useful action by continuously collecting training data in the background while the model is being used. The training data collection module operates continuously to gather user behavior data, and the model training module periodically retrains the model using this accumulated data, ensuring both stability and continuous adaptation to user behavior changes
Solution Approach 2:
The system implements dynamic training where the model can switch between using pre-trained weights and updating with new data based on performance thresholds. When the model performs well, it maintains stability; when performance degrades, it dynamically retrains with new data, achieving both stability and adaptability
Data Source
AI summary
A training method and a training system for a machine learning system are provided. The method includes allocating training data to a plurality of working machines; dividing training data allocated by each working machine into a plurality of data pieces; obtaining a local weight and a local loss function value calculated by each working machine based on each data piece; aggregating the local weight and the local loss function value calculated by each work machine based on each data piece to obtain a current weight and a current loss function value; performing model abnormality detection using the current weight and/or the current loss function value; inputting a weight and a loss function value of a previous aggregation to the machine learning system for training in response to a result of the model abnormality detection being a first type of abnormality; and modifying the current weight and/or the current loss function value to a current weight and/or a current loss function value within a first threshold in response to the result of the model abnormality detection being a second type of abnormality, and inputting thereof to the machine learning system for training.