M$^3$Eval: Multi-Modal Memory Evaluation through Cognitively-Grounded Video Tasks
2026-06-03 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial IntelligenceComputation and Language
AI summaryⓘ
The authors study how well multi-modal AI models remember information from long videos. They created a new test called M³Eval to check different memory abilities in these models, inspired by how human memory works. Their experiments show that these models often mix up information from different video streams, remember locations better than timing, and have trouble with symbolic memory. This work helps researchers understand and improve memory in AI that processes video and other types of data together.
multi-modal modelslong-form video understandingmemory evaluationM³Eval benchmarkcognitive psychologydisentangled representationsspatial memorytemporal memorysymbolic memory
Authors
Jie Huang, Ruixun Liu, Sirui Sun, Xinyi Yang, Yin Li, Yixin Zhu, Yiwu Zhong
Abstract
As multi-modal models advance towards long-form video understanding, memory emerges as a critical capability. Despite substantial efforts in developing video datasets and benchmarks, existing works primarily focus on perception and reasoning, without systematically evaluating memory: what models retain, how faithfully information is preserved, and how robust memory remains under interference. To address this gap, we introduce M$^3$Eval, the first comprehensive evaluation framework and benchmark for probing different memory dimensions in multi-modal models. Grounded in cognitive psychology, our design features carefully constructed tasks that isolate key aspects of memory. Leveraging M$^3$Eval, we conduct extensive experiments across representative multi-modal models, revealing consistent weaknesses and distinctive behaviors. We find that models struggle to maintain disentangled representations when processing parallel video streams, exhibit interference patterns differing substantially from those observed in human memory, ground memory sources more reliably in the spatial domain than the temporal domain, and demonstrate limited symbolic memory. Collectively, our benchmark provides a valuable resource for future research, while our findings highlight memory as a fundamental yet underexplored capability and offer insights for designing more effective memory mechanisms in multi-modal models. Our code and dataset are available at https://pku-value-lab.github.io/m3eval-homepage.