Research map / Label-efficient and robust learning / Domain adaptation and generalization
Improving Non-Transferable Representation Learning by Harnessing Content and Style
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a non-transferable representation learning method that separates content and style to restrict generalization to target domains, improving over statistical dependence-reduction approaches.
Problem
Non-transferable learning (NTL) aims to restrict model generalization toward target domain(s); existing statistical methods reduce source-target dependence but fail to distinguish styles from contents, leading to spurious correlations between style and labels.
Approach
Explicitly distinguishes content from style in representations so that non-transferability is induced by harnessing both, rather than relying on statistical dependence reduction that conflates styles and contents.
Experimental setup
Baselines: statistical dependence reduction NTL methods
Builds on: non-transferable learning
Closest papers by meaning
- Towards the Causal Complete Cause of Multi-Modal Representation Learning (ICML 2025 · similarity 0.86)
- Content-Style Learning from Unaligned Domains: Identifiability under Unknown Latent Dimensions (ICLR 2025 · similarity 0.85)
- Your Transferability Barrier is Fragile: Free-Lunch for Transferring the Non-Transferable Learning (CVPR 2024 · similarity 0.85)
- Non-negative Contrastive Learning (ICLR 2024 · similarity 0.85)
- Sample-efficient Learning of Concepts with Theoretical Guarantees: from Data to Concepts without Interventions (NeurIPS 2025 · similarity 0.85)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.