- Rating
- 3
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 3
Read review
Paper summary
This paper proposes CALM, a training-free KV-cache management framework for streaming video understanding. The method uses leave-one-out (LOO) output divergence as a diagnostic reference for token importance and introduces a gradient-based KV Taylor score computed through a single probe forward–backward pass. CALM combines this score with depth-dependent recency weighting, a center-weighted allocation of cache capacity, conditional novelty reranking, and a bounded text memory containing generated chunk summaries.
Critical strengths
- The paper investigates the scoring signal rather than relying only on end-to-end accuracy.
- Experiments demonstrate consistent improvements across multiple backbones.
- Bounded-memory design and update-cost breakdown address practical concerns.
Critical weaknesses
- The relationship between KV Taylor and actual token deletion remains indirect.
- Depth-profile analysis does not directly justify the proposed allocation of capacity.
- The magnitude and robustness of the improvements need more precise presentation.
Questions for authors
- How sensitive are the rankings and downstream results to guidance-prompt wording and the use of a single self-predicted token?
- Can the authors demonstrate that layers with stronger KV Taylor–LOO agreement also benefit more from additional cache capacity?
- Can the authors provide paired uncertainty estimates for comparisons with HERMES and the attention-based CALM variant? Do the conclusions remain consistent across cache budgets other than 4K?