Researchers have developed MoME, a novel framework designed to enable privacy compliance in re-identification models. This switchable system allows a single pre-trained backbone to operate in two modes: 'Full' for standard inference and 'Core' for unlearned inference. MoME achieves this by separating core parameters that retain retrieval capabilities from parasitic parameters that store full-data knowledge, ensuring the core parameters are not affected by data deletion requests. This approach allows for selective forgetting of individuals while maintaining the model's ability to recognize others, with performance validated on datasets like Market-1501, VeRi-776, and RegDB. AI
IMPACT This framework could enhance privacy compliance in AI systems by allowing selective data deletion without full retraining.
RANK_REASON The cluster contains a research paper detailing a new framework for AI model unlearning. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
