Machine Learning Method for New Class Discovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods face challenges in efficiently utilizing unlabeled data for learning due to the scarcity of labeled data, and they often rely on incorrect model assumptions, leading to suboptimal performance, especially in applications where new classes may not be known in advance.
Innovation Solution
An integrated approach that combines semi-supervised learning and active learning, iteratively refining models using both labeled and unlabeled data, requesting labels for high-confidence data, and incorporating a new class discovery module to identify and verify novel classes, ensuring conservative model assumptions and labeler feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If semi-supervised learning is used to utilize unlabeled data, then the model can learn from more data, but wrong model assumptions lead to learning wrong labels and worse performance
Solution Approach 1:
The patent implements a feedback mechanism where the active learning component queries the semi-supervised model's confidence predictions and uses labeler feedback on high-confidence predictions to iteratively refine the model. This feedback loop allows the system to correct wrong model assumptions by incorporating human verification, thereby maintaining reliability while utilizing large amounts of unlabeled data through the semi-supervised learning component.
Solution Approach 2:
The system applies partial action by selectively querying only high-confidence predictions from the semi-supervised model for labeler verification, rather than verifying all predictions. This partial verification approach allows the system to benefit from the semi-supervised model's ability to process large amounts of unlabeled data while maintaining reliability through targeted human feedback on the most certain predictions.
2Productivity
If active learning is used to select informative query instances, then training set size is reduced, but practical realities violate basic assumptions and query instances lack semantic meaning
Solution Approach 1:
The patent introduces the semi-supervised learning model as an intermediary between the active learning component and the labeler. The semi-supervised model processes unlabeled data and provides confidence-based filtering, acting as a mediator that identifies high-quality query instances with semantic meaning. This intermediary approach allows active learning to maintain productivity by selecting only the most informative instances while ensuring data quality through the semi-supervised model's confidence assessment.
Solution Approach 2:
The system performs preliminary action by using the semi-supervised learning model to pre-process unlabeled data and identify high-confidence predictions before submitting them for labeler verification. This preliminary filtering ensures that only data instances with semantic meaning and high confidence scores are queried, maintaining ease of operation while preserving learning efficiency through targeted selection.
3Adaptability or versatility
If new class discovery is performed on unlabeled data, then novel classes can be identified, but the process becomes more complex
Solution Approach 1:
The patent merges new class discovery with the existing semi-supervised learning and active learning framework. The new class discovery component is integrated into the iterative loop, where it identifies novel classes from unlabeled data and feeds them back to update the semi-supervised model. This merging approach allows the system to gain adaptability through new class discovery while managing complexity by incorporating the functionality into the existing unified framework rather than adding a separate complex system.
Data Source
AI summary
A computerized efficient machine learning method for classification of data and new class discovery inputs labeled data and unlabeled data into a computer memory for a computerized machine tool to perform (a) initial supervised learning using the labeled data to generate a classifier, (b) semi-supervised learning using the labeled data, the classifier and the unlabeled data to generate an updated classifier and high confidence data, (c) active learning using the updated classifier and the unlabeled data to generate a data label request and receive new class labeled data to generate augmented labeled data, (d) new class discovery using the updated classifier and the data label request to generate data of potential new classes and receive labels for potential new class data to generate new class labeled data, and (e) supervised learning using the high confidence data, the labeled data and the augmented labeled data to generate an output classifier.