Machine Learning Model Updates for Stable Data Drift Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for updating models in non-stationary environments face challenges in achieving both robustness and accuracy, particularly in handling abrupt and gradual changes, leading to deteriorated model accuracy and increased maintenance burdens.
Innovation Solution
A method that adopts a passive approach to follow gradual changes, automatically detecting when to update the model, and uses a cost function with regularization terms to minimize parameter changes, ensuring stability and accuracy by only updating when data trends change.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If model update frequency is increased to improve followability of data trends, then model accuracy improves, but maintenance burden and computational cost increase
Solution Approach 1:
The patent implements a feedback mechanism by calculating the difference between prediction values from the old model and actual target values, using this feedback to determine when model updates are necessary. This allows the system to maintain accuracy by updating only when performance degradation is detected, rather than following a fixed update schedule.
Solution Approach 2:
The patent makes the model update process dynamic by adapting the update frequency based on actual data drift conditions. The system transitions from static periodic updates to dynamic updates triggered by performance metrics, allowing the model to respond to changing conditions only when necessary.
2Adaptability or versatility
If model parameters are updated frequently to track data trends, then followability improves, but model stability and robustness deteriorate
Solution Approach 1:
The system uses feedback from performance evaluation to control parameter updates. By monitoring the difference between old model predictions and actual values, the system only updates parameters when necessary, preventing unnecessary fluctuations that would compromise stability while still adapting to genuine data drift.
Solution Approach 2:
The patent carefully controls parameter changes by using regularization terms in the cost function that penalize large parameter deviations. This allows the model to adapt to data trends through controlled parameter adjustments rather than abrupt changes, maintaining stability while achieving followability.
3Stability of the object's composition
If regularization terms are added to minimize parameter changes, then model stability improves, but adaptability to new data trends may be reduced
Solution Approach 1:
The patent uses regularization terms in the cost function to control parameter changes during updates. The regularization component penalizes large deviations from previous parameter values, ensuring stable updates that maintain model robustness while still allowing necessary adaptations to data drift through controlled parameter adjustments.
Data Source
AI summary
An information processing apparatus of an embodiment includes one or more hardware processors. The one or more hardware processors receive input of parameter of a model to be estimated by machine learning and input of first input data. The one or more hardware processors train, by using the first input data as training data, the model using a cost function for which a cost is smaller as a change in the parameter is smaller.