VideoStreaming
Streaming Long Video Understanding with Large Language Models
Rui Qian, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Shuangrui Ding, Dahua Lin, et al. · 2024-05-25
- Management
- T2 Dynamic Memory
- Memory changes during deployment through a stable management policy.
- Representation
- Internalized latent & parametric
How experience becomes memory
Each path is one memory component, traced from what is stored to how it is found again.
propagated video memory
- Represented asPersistent multimodal latent stateInternalized latent & parametric
- Stored inFlat sequential records
- Retrieved bySimilarity search, Latent / parametric activation
- Source evidenceSource not retained
Each clip is folded into constant-size historical latent memory.Full-paper method / memory representation and retrieval · PDF p. 2 · Figure 1
How memory changes (T2 operations)
Update · Learned update module
A fixed learned update propagates memory across clips. For VideoStreaming, this specifically supports a trained but frozen update module of propagated video memory.
Related systems
- FlexMemT2 · Video & streaming
- Latent Spatial MemoryT2 · Video & streaming
- Light-OmniT2 · Video & streaming
- MA-LMMT2 · Video & streaming
- MemLearnerT2 · Video & streaming
- Memorize-and-GenerateT2 · Video & streaming