New CAZO method enhances memory-efficient test-time adaptation

Researchers have developed a new zeroth-order optimization method called Curvature-Aware Zeroth-Order Optimization (CAZO) for memory-efficient test-time adaptation (TTA). This method aims to improve the performance of pre-trained models on new datasets without requiring backpropagation, making it suitable for on-device applications. CAZO leverages the observation of a persistent low-rank Hessian structure in the loss function during adaptation to construct a covariance matrix for more efficient gradient estimation. Experiments show that CAZO outperforms existing TTA methods in accuracy and memory efficiency. AI

IMPACT This new method could enable more efficient on-device adaptation of AI models, reducing memory overhead and complexity.

RANK_REASON The cluster contains a research paper detailing a new optimization method for machine learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• CORE Recommender
• Curvature-Aware Zeroth-Order Optimization for Memory-Efficient Test-Time Adaptation
• DagsHub
• Gotit.pub
• Hugging Face
• ScienceCast
• Test-Time Adaptation
• Zeroth-order optimization with orthogonal random directions

• paper
• other

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