mlsanityData to 14 Oct 2026

Who Speaks for the Pruned? Visual Token Pruning as Coverage Optimization

arXiv 2609.03158 ↗First posted 2 Sep 2026cs.CV cs.CL cs.LGcited by 2

On the map: Multimodal LLMs / vision-language models → Visual token pruning and efficiency

also in 1 more
Abstract (arXiv)

Visual token pruning reduces the inference cost of vision-language models (VLMs), but most methods only ask which tokens to keep. This retained-token view can keep redundant high-scoring tokens while leaving discarded evidence without a close representative. We propose CoverPruner, a training-free pruner that asks the complementary demand-side question: after a token is removed, which surviving original token represents it for the target VLM? CoverPruner formulates pruning as Representational Coverage Maximization (RCM), covering the full projected visual-token set with query-weighted demand. It instantiates RCM with projector-space coverage and a lightweight first-layer attention probe. Across multiple VLM architectures and compression rates, CoverPruner achieves the best average accuracy among all compared methods, with the largest gains usually appearing under aggressive compression.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 16 of its 34 references that are in the map. First posted on arXiv 2 Sep 2026.

Builds on

16 of 34 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025MMTok: Multimodal Coverage Maximization for Efficient Inference of VLMs25 Aug
  2. 2025Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs12 Jun
  3. 2025Similarity-Aware Token Pruning: Your VLM but Faster14 Mar
  4. 2025Stop Looking for Important Tokens in Multimodal Language Models: Duplication Matters More17 Feb
  5. 2024PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction22 Oct

Other references · 11

  1. 2023Improved Baselines with Visual Instruction Tuningcited by 4,418
  2. 2024LLaVA-OneVision: Easy Visual Task Transfercited by 2,719
  3. 2016Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answeringcited by 2,516
  4. 2023MMBench: Is Your Multi-modal Model an All-around Player?cited by 2,141
  5. 2022Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answeringcited by 1,919
  6. 2019Towards VQA Models That Can Readcited by 1,718
  7. 2022ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoningcited by 1,479
  8. 2022Token Merging: Your ViT But Fastercited by 1,021
  9. 2016A Diagram Is Worth A Dozen Imagescited by 804
  10. 2024LLaVA-Video: Video Instruction Tuning With Synthetic Datacited by 583
  11. 2025Why 1 + 1 < 1 in Visual Token Pruning: Beyond Naive Integration via Multi-Objective Balanced Coveringcited by 11

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026CoViST: Visual Token Compression via Composable Statescited by 0
  2. 2026MiCo: Mutual Information Coverage Optimization through Semantic Erasure Modeling for Efficient MLLM Inferencecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026ACPruner: Visual Token Pruning as Biased Attention Coverage Maximization in LVLMs
  2. 2026QCPruner: Query-Conditioned Population Coverage for Visual Token Pruning
  3. 2026Beyond Surrogate Gradients: Fully Differentiable Token Pruning for Vision-Language Models
  4. 2026SwiftVLM: Efficient Vision-Language Model Inference via Cross-Layer Token Bypass
  5. 2025A Glimpse to Compress: Dynamic Visual Token Pruning for Large Vision-Language Models
  6. 2026The Model Knows Which Tokens Matter: Automatic Token Selection via Noise Gating
  7. 2026VPRune: Efficient Training-free Pre-LLM Visual Token Pruning
  8. 2026OccamToken: Efficient VLM Inference with Training-Free and Budget-Adaptive Token Pruning