Write Once, Query Many: Evolving Compact Reusable Memory for Long Videos

Key points

• Long videos often exceed the context capacity of multimodal large language models (MLLMs), while agentic search repeats costly video processing and grounding for each query.1
• Reusable memory can amortize this cost, but must preserve information for questions unknown during construction.1
• This representation combines compact free-form text with sparse images to preserve semantic content and complementary visual details across temporal scales, supporting temporal navigation without a fixed semantic schema.1
• With the underlying models frozen, we evolve their workflows and operators using distinct feedback: query-independent source fidelity and coverage for the Writer, and QA failures in evidence localization and utilization for the Reader.1

Sources (1)

• [1]Write Once, Query Many: Evolving Compact Reusable Memory for Long Videos
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 07:58 AM

  We propose Video-EMem, an evolvable video memory harness built on time-anchored perceptual observations.

  Long videos often exceed the context capacity of multimodal large language models (MLLMs), while agentic search repeats costly video processing and grounding for each query.

Extractive summary: sentences quoted from the sources.