Grounded perceptual memory for personalized agents: VLM grounding, specialist identity encoders, and a training-free in-model attention bank. Paper: arXiv:2608.28609
reproducible-research transformers pytorch personalization attention face-recognition arxiv speaker-recognition multimodal qwen llm-agents vision-language-models user-memory parametric-memory multimodal-memory perceptual-memory
-
Updated
Sep 1, 2026 - Python