Unsupervised Pattern Detection via Activation Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing unsupervised learning methods for detecting repeating patterns in data sequences, such as parts of objects in images, require prior annotation or information about object categories, and often lack transparency and independence in evaluating detector performance.
Innovation Solution
A novel unsupervised learning method that uses pre-trained deep convolutional neural networks to construct detectors of parts of objects as weighted sums of activation maps, without requiring prior annotation or knowledge of object categories. The method employs cost functions to ensure locality, uniqueness, and grouping properties in the detection of connected parts of objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning methods are used for detecting repeating patterns, then detection accuracy is improved, but annotation cost and complexity increase significantly
Solution Approach 1:
The system performs self-annotation by automatically detecting repeating patterns in unannotated data through unsupervised learning. The detector identifies patterns without human intervention, and the system generates its own training labels by detecting repetitions across multiple data instances, eliminating the need for external annotation services
Solution Approach 2:
The system performs preliminary detection of repeating patterns across the entire dataset before final classification. By pre-identifying repeating structures and using them as features for subsequent supervised learning, the system reduces the annotation burden while maintaining accuracy
2Ease of manufacture
If crowdsource annotation is used to reduce costs, then annotation cost decreases, but annotation quality and training performance deteriorate
Solution Approach 1:
Instead of relying on crowdsource annotators, the system performs self-annotation by automatically identifying repeating patterns in unannotated data. The unsupervised detector generates reliable labels through pattern repetition detection, eliminating both the cost and quality issues associated with crowdsource annotation
Solution Approach 2:
The system introduces an intermediate unsupervised pattern detection step between raw data and supervised learning. This intermediary detector processes unannotated data to generate pattern-based features and pseudo-labels, which then serve as input for the supervised classifier, bridging the gap between unannotated data and training requirements
3Device complexity
If semi-supervised methods are used, then annotation requirement decreases, but detection coverage and reliability are limited
Solution Approach 1:
The system performs preliminary unsupervised detection of repeating patterns across the entire dataset before final classification. By pre-identifying all repeating structures regardless of their relevance to the final task, the system ensures comprehensive coverage and reliability while minimizing annotation requirements
Solution Approach 2:
The unsupervised pattern detector serves multiple functions: it identifies repeating patterns for use as features, generates pseudo-labels for training, and provides pattern-based features that are independent of any specific classification task. This multi-functionality increases both coverage and reliability while reducing annotation needs
4Measurement precision
If deep convolutional neural networks are used for object recognition, then recognition effectiveness is improved, but interpretability and transparency decrease
Solution Approach 1:
The system segments the recognition process into two interpretable components: (1) unsupervised detection of repeating patterns that provides transparent pattern-based features, and (2) supervised classification using these explicit features. This segmentation maintains recognition effectiveness while improving interpretability by making the decision process more transparent and traceable
Data Source
AI summary
A method, implemented by computer, is provided for unsupervised training of a model for detecting repeating patterns in a dataset, the model being composed of a detection layer including at least: a detector of repeating patterns configured to receive a set of pertinent features extracted from the data and supply as output an activation map composed of a set of activation scores, an activation layer consisting at least in normalizing the activation maps, the learning of each detector being limited to the observance of a locality criterion consisting in maximizing a region of the activation map.