Auxiliary Model Selects Diverse Training Data for Limited-Capacity ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Active machine learning requires high-quality labeled training examples, which are costly to obtain, especially for limited-capacity machine learning models, as they struggle to differentiate between distinct observations, leading to 'color-blindness' and inefficiencies in training.
Innovation Solution
An auxiliary machine learning model with a larger capacity is used to identify the scope of 'color-blindness' in a target model, producing new high-quality labeled examples to incrementally feature or refine the target model, enhancing diversity and efficiency in training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If many users are employed to interpret unlabeled documents to determine viability for machine learning purposes, then the quality of labeled training examples is improved, but the cost exceeds desired targets
Solution Approach 1:
An auxiliary machine learning model with larger capacity is introduced as an intermediary to pre-process and identify high-quality unlabeled documents. This auxiliary model acts as a mediator between the vast pool of unlabeled documents and the limited-capacity target model, filtering and selecting only the most valuable training examples. This intermediary layer reduces the burden on human annotators and lowers the cost of obtaining high-quality labeled training data while maintaining or improving the quality of training examples for the target model.
2Device complexity
If the scope of color-blindness in a machine learning model is not addressed, then the device complexity is reduced, but the productivity of training is worsened due to inefficiencies
Solution Approach 1:
The training process is segmented into multiple stages: first, an auxiliary model with larger capacity identifies the scope of color-blindness and selects diverse high-quality unlabeled documents; second, these selected documents are used to incrementally feature the limited-capacity target model. This segmentation allows the system to address the color-blindness issue efficiently by breaking down the complex training process into manageable steps, improving training productivity without excessively increasing the complexity of the final target model.
Solution Approach 2:
The auxiliary machine learning model performs preliminary action by pre-identifying and selecting high-quality unlabeled documents before they are used to train the target model. This preliminary filtering and selection process ensures that only the most valuable and diverse training examples are passed to the target model, thereby improving training efficiency and addressing color-blindness issues before the main training process begins, without adding significant complexity to the target model architecture.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Technologies are described herein for active machine learning. An active machine learning method can include initiating active machine learning through an active machine learning system configured to train an auxiliary machine learning model to produce at least one new labeled observation, refining a capacity of a target machine learning model based on the active machine learning, and retraining the auxiliary machine learning model with the at least one new labeled observation subsequent to refining the capacity of the target machine learning model. Additionally, the target machine learning model is a limited-capacity machine learning model according to the description provided herein.