paper-bot: paper-bot/2026-w30

[github-actions]

paper-bot batch

2 papers from top systems venues, ranked by citation count. Each paper is its own commit so reviews are surgical.

• 26ae0f7 (520 cites, Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication 2018) — Chameleon: scalable adaptation of video analytics
  • file: ai/chameleon/26ae0f7-chameleon-scalable-adaptation-of-video-analytics.md (ba1c1a3)
  • tl;dr: Chameleon is a controller that dynamically selects the best NN configuration (e.g., resolution, frame rate) for video analytics pipelines, exploiting temporal and spatial correlations to amortize search costs and achieve 2-3x resource savings or 20-50% accuracy gains.
• 2205.05198 (491 cites, Conference on Machine Learning and Systems 2022) — Reducing Activation Recomputation in Large Transformer Models
  • file: gpu-training/misc/2205.05198-reducing-activation-recomputation-in-large-transformer-models.md (707b368)
  • tl;dr: This paper reduces activation recomputation overhead in large transformer training by introducing sequence parallelism and selective activation recomputation, achieving 5x memory reduction and over 90% less recomputation overhead.

How to review

Comment on this PR with one of:

• /refine <id> <free-form instructions> — re-write the analysis with your feedback. Folds into the original commit.
• /regenerate <id> — re-fetch the paper and run the full analyzer again.
• /reject <id> — revert the paper's commit.

<id> can be an arXiv id (when shown), the 7-char paper-id, or the file path. Examples:

/refine 2401.09670 把 evaluation 段补充和 vLLM 的对比数字
/regenerate 72f77a3
/reject gpu-training/misc/2024-some-paper.md

When done, merge with Squash & merge to keep main history clean.