New TTA method uses ZOO and model merging for resource-limited devices

Researchers have developed a new method for test-time adaptation (TTA) that addresses the resource limitations of edge devices. By integrating zeroth-order optimization (ZOO) with model merging within a cross-device collaborative TTA framework, their approach bypasses the need for computationally intensive backpropagation. This method is designed for systems with a mix of resource-rich and resource-limited devices, enabling model updates on the latter by using only forward processing. The technique also incorporates a preprocessing strategy to trim non-influential weights and reduce inter-model redundancy, enhancing the synergy between ZOO and model merging. AI

IMPACT Enables on-device model adaptation, potentially improving performance and personalization for edge AI applications.

RANK_REASON The cluster contains a single arXiv paper detailing a new method for test-time adaptation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• BP
• CDC-TTA
• Cross-device collaborative Test-time Adaptation
• Model Merging for Linguistic Transfer
• Test-Time Adaptation
• Zeroth-order optimization with orthogonal random directions
• Zootaxa

• paper
• infra

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