ComMem Dual-Memory System for Vision-Language Model Test-Time Adaptation

ComMem introduces a dual-memory architecture for vision-language model adaptation using a fast-adapting visual cache and a slow-integrating abstract textual memory. This approach enables continuous knowledge accumulation during test-time deployment by mimicking hippocampal and neocortical memory functions.