New theory explains why Zeroth-Order adaptation reduces model forgetting

Researchers have developed a new theoretical framework, Randomized Shaping Theory, to explain why Zeroth-Order (ZO) adaptation methods in continual learning may lead to less forgetting than first-order (FO) methods. The theory suggests that ZO adaptation, when properly analyzed, can preserve more previously acquired knowledge by selectively contracting anisotropic components of adaptation. This theoretical insight has led to a new algorithm called RISE, which applies calibrated ZO shaping to exact FO gradients within parameter blocks to improve the stability-plasticity tradeoff in continual learning. AI

IMPACT Introduces a theoretical explanation for improved continual learning, potentially leading to more robust AI systems that retain knowledge over time.

RANK_REASON The cluster contains a new academic paper detailing a theoretical framework and a proposed algorithm for continual learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Continual learning
• First-Order adaptation
• Randomized Shaping Theory
• RISE
• Zeroth-Order adaptation

• paper
• safety

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