Concept-Based Few-Shot Learning Without Prior Knowledge
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing few-shot learning methods require prior knowledge of attribute text for class classification, limiting their effectiveness in tasks with small amounts of data.
Innovation Solution
A concept-based few-shot learning method that estimates task embeddings from support data, calculates slot probabilities for a concept memory, extracts features from query and support data, and generates synthesis features with maximum similarity to execute tasks without prior knowledge, using a concept memory, task estimation, feature extraction, and synthesis feature generation units.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If prior knowledge of attribute text is used for class classification, then classification accuracy is improved, but the requirement for prior knowledge increases system complexity and limits applicability
Solution Approach 1:
The system extracts concepts automatically from support data without requiring external prior knowledge. The concept extraction unit identifies relevant concepts directly from the input data, enabling the system to serve itself by generating necessary knowledge representations during the learning process rather than relying on pre-defined attribute texts
Solution Approach 2:
A concept memory is introduced as an intermediary component that stores extracted concepts and their features. This concept memory acts as a bridge between support data and query data, allowing the system to perform classification by comparing features against stored concepts without requiring direct prior knowledge of attribute texts
2Reliability
If a concept memory is introduced to store concept features, then few-shot learning performance is improved, but the device structure becomes more complex
Solution Approach 1:
The concept memory is pre-populated with concept features extracted from base data before the few-shot learning task begins. This preliminary action of storing concept representations in advance allows the system to quickly perform few-shot learning by comparing query features against pre-stored concepts, improving performance without adding complex real-time processing structures
Solution Approach 2:
The system changes the parameter representation by storing concept features in a structured format within the concept memory. Instead of working with raw attribute texts, the system transforms concepts into feature vectors that can be efficiently compared and processed, improving few-shot learning performance while maintaining manageable system complexity through parameter transformation
Data Source
AI summary
A concept based few-shot learning method is disclosed. The method includes estimating a task embedding corresponding to a task to be executed from support data that is a small amount of learning data; calculating a slot probability of a concept memory necessary for a task based on the task embedding; extracting features of query data that is test data, and of the support data; comparing local features for the extracted features with slots of a concept memory to extract a concept, and generating synthesis features to have maximum similarity to the extracted features through the slots of the concept memory; and calculating a task execution result from the synthesis feature and the extracted concept by applying the slot probability as a weight.