New framework unifies self-ensembling for test-time prompt tuning

Researchers have introduced USE, a unified self-ensembling framework designed to enhance test-time adaptation for vision-language models like CLIP. This framework interprets test-time prompt tuning as learning from self-generated pseudo-labels, ensuring consistency between optimization and inference stages. By adaptively emphasizing the test image over its augmented views, USE obtains more reliable pseudo-labels and demonstrates superior performance across various datasets compared to existing methods. Additionally, a simplified self-ensembling strategy, SE, can function as a lightweight, optimization-free test-time adaptation technique. AI

IMPACT This research could lead to more robust and accurate performance of vision-language models on unseen data by improving test-time adaptation techniques.

RANK_REASON The cluster contains an academic paper detailing a new framework and methodology for test-time adaptation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• DagsHub
• Hugging Face
• Self ensembling techniques for generating magnetic resonance images from spatial frequency data
• Test-Time Adaptation
• Test-Time Prompt Tuning
• USE

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →