Domain Adapted Class Means Classifier Using Stacked Marginalized Denoising Autoencoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing domain adaptation methods require labeled target data and abundant source data, which is often not available in practice due to confidentiality issues, leading to poor performance when applying models from source domains to target domains.
Innovation Solution
The method employs unsupervised Stacked Marginalized Denoising Autoencoders (sMDA) to adapt class means from source domains to the target domain, allowing for domain adaptation without direct access to source data by using precomputed class representations and deep learning techniques, enabling feature extraction and prediction of class labels in the target domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If domain adaptation methods use labeled target data and abundant source data, then classification performance is improved, but data availability and privacy requirements worsen
Solution Approach 1:
The patent extracts only the essential information (class means) from source domains while leaving the actual source data behind. This allows domain adaptation to proceed with compact representations rather than requiring access to the full source datasets, resolving the contradiction between needing source data for performance and the unavailability of such data due to privacy or storage constraints
Solution Approach 2:
The patent creates simplified copies of source domain data in the form of class means (compact statistical representations). These copies capture the essential characteristics needed for domain adaptation without requiring the original detailed data, enabling performance comparable to using full source data while avoiding data availability issues
2Measurement precision
If domain adaptation methods access original source samples, then adaptation accuracy is improved, but privacy preservation and data security worsen
Solution Approach 1:
The patent extracts only aggregated statistical information (class means) from source domains, removing all identifiable and contextual information that would compromise privacy. This extraction approach maintains adaptation accuracy while eliminating privacy risks associated with accessing original source samples
Solution Approach 2:
The patent uses disposable, anonymized representations (class means) instead of valuable, sensitive original data. These simplified representations serve the adaptation purpose effectively but contain no sensitive information that could be misused, resolving the privacy concern
3Adaptability or versatility
If domain adaptation uses multiple source domains, then robustness and versatility are improved, but system complexity and computational requirements worsen
Solution Approach 1:
The patent extracts essential characteristics from multiple source domains into compact class mean representations. This extraction simplifies the integration of multiple sources by reducing each source to its fundamental statistical properties, making multi-source adaptation feasible without overwhelming complexity
Solution Approach 2:
The patent merges information from multiple source domains by computing class means that aggregate statistics across sources. This merging process consolidates diverse source information into unified representations that can be efficiently processed, reducing the complexity of handling multiple separate source domains
Data Source
AI summary
A domain-adapted classification system and method are disclosed. The method includes mapping an input set of representations to generate an output set of representations, using a learned transformation. The input set of representations includes a set of target samples from a target domain. The input set also includes, for each of a plurality of source domains, a class representation for each of a plurality of classes. The class representations are representative of a respective set of source samples from the respective source domain labeled with a respective class. The output set of representations includes an adapted representation of each of the target samples and an adapted class representation for each of the classes for each of the source domains. A class label is predicted for at least one of the target samples based on the output set of representations and information based on the predicted class label is output.