Labeled Image Subset Selection for Cross-Domain Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning systems rely on full annotated datasets for training, leading to substantial computational overhead and inefficiencies in adapting to new or shifting target domains, resulting in inaccurate classifications and excessive false positives in real-world deployment scenarios.
Innovation Solution
A system that performs domain-aligned selection of labeled images by clustering a source dataset and ranking the resulting clusters based on feature-level similarity to a target dataset, enabling efficient construction of a refined training subset that enhances the performance of downstream visual classification models without requiring full retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If full annotated datasets are used for training, then model accuracy is maintained, but computational overhead increases substantially
Solution Approach 1:
The patent extracts and removes irrelevant or redundant images from the source dataset by comparing feature representations against target domain images. Only the most relevant subset of labeled images is retained for training, thereby reducing computational overhead while preserving the accuracy needed for target domain classification
Solution Approach 2:
The patent changes the parameter of dataset size by selecting a refined subset of labeled images based on similarity metrics. This parameter change reduces the volume of training data from the full source dataset to only those images most relevant to the target domain, decreasing computational requirements while maintaining performance
2Reliability
If conventional machine learning systems use full annotated datasets, then comprehensive training is achieved, but adaptation to new domains becomes inefficient
Solution Approach 1:
The patent applies local quality by selecting source images whose feature representations locally match the target domain characteristics. Rather than uniformly using all source data, it identifies and retains only those local subsets (individual images or small groups) that exhibit visual characteristics similar to the target domain, enabling efficient adaptation
Solution Approach 2:
The patent performs preliminary action by pre-comparing source dataset feature representations with target domain feature representations before training. This preliminary similarity assessment identifies and selects the most relevant source images in advance, so that when adaptation to new domains is needed, the model can quickly utilize pre-filtered relevant data without reprocessing the entire source dataset
3Reliability
If full datasets are used for training, then model robustness is improved, but training time and computational resources increase
Solution Approach 1:
The patent extracts only the essential training images needed for robust performance by comparing feature representations and selecting those with highest similarity to target domain. This extraction of the critical subset maintains model robustness while eliminating unnecessary training time spent on irrelevant images
Solution Approach 2:
The patent applies partial action by using only the necessary portion of the source dataset rather than the complete set. By selecting a refined subset of labeled images that are most relevant to the target domain, it achieves sufficient training coverage for robustness without the excessive time cost of processing the entire source dataset
Data Source
AI summary
An example operation may include at least one of receiving, from a source dataset, a plurality of labeled images, receiving, from a target dataset, a plurality of images associated with a different visual domain, extracting, from each of the plurality of labeled images and each of the plurality of images from the target dataset, one or more feature representations indicative of visual characteristics, grouping the plurality of labeled images into a plurality of image clusters based on similarity among the one or more feature representations, comparing the one or more feature representations of each of the plurality of image clusters to the one or more feature representations of the plurality of images from the target dataset to determine a similarity ranking for each image cluster, selecting, from the plurality of image clusters, a subset of labeled images based on the similarity ranking and a selection limit, and storing the subset of labeled images in a memory.