Proxy-Model Active Learning for Efficient Data Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional active learning methods face inefficiencies due to repeated calculations of utility scores for data points, especially in large datasets, leading to computational bottlenecks and decreased model training efficiency.
Innovation Solution
A method using a proxy model to estimate the utility distribution of a raw data pool, determining selection and exclusion criteria based on this distribution, and iteratively selecting data for labeling to reduce the need for repeated utility score calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional active learning methods recalculate utility scores for data points in each batch, then data selection can be based on current model performance, but computational resource consumption increases significantly
Solution Approach 1:
The patent pre-calculates and stores utility scores for all data points before the active learning process begins. These pre-computed utility scores are then used during the iterative batches without recalculating, thereby maintaining data selection accuracy while dramatically reducing computational resource consumption during training.
2Adaptability or versatility
If utility scores are recalculated in each batch for large datasets, then the model can adapt to new data, but the training process becomes a computational bottleneck
Solution Approach 1:
Utility scores are pre-computed for all data points before training begins. During iterative batches, the system selects data based on these pre-computed scores and model performance, avoiding the need to recalculate utility scores for each batch while maintaining the ability to adapt to new data through iterative model training.
Solution Approach 2:
The patent creates a copy of the utility score calculations that can be reused across multiple batches. Instead of performing the full utility score calculation each time, the system reuses the pre-computed utility scores, effectively copying the computational work from the initial setup phase to subsequent training phases.
3Measurement precision
If repeated utility score calculations are performed for every new batch, then data selection remains accurate, but computational cost becomes extremely high
Solution Approach 1:
The system performs the computationally intensive utility score calculations once during an initial phase before training begins. These accurate utility scores are then stored and reused for data selection across multiple batches, maintaining measurement precision while eliminating the need to repeat the expensive calculations in each subsequent batch.
Data Source
AI summary
A computer-implemented method for active learning is provided. The method includes steps of using a proxy model to estimate a utility distribution of a raw data pool based on a raw data subset obtained from the raw data pool, determining a selection criterion based on the utility distribution, performing a data selection process based on the selection criterion, and using the training data pool to train the target model. The data selection process involves steps of using the proxy model to calculate a utility score associated with a raw image from the raw data pool, using the selection criterion to selectively provide the raw image to an oracle to obtain a selected image corresponding to the raw image, and incorporating the selected image into the training data pool. The utility score associated with the raw image provided to the oracle meets the selection criterion.