通过覆盖优化提升视觉标记剪枝效果，更精准保留关键信息。
Who Speaks for the Pruned? Visual Token Pruning as Coverage Optimization
- 将剪枝问题转化为表示覆盖最大化，关注被删标记的替代代表
- 在多种模型和压缩率下平均准确率领先，激进压缩时提升最显著
- 无需训练，利用轻量级注意力探针实现高效覆盖评估
视觉标记剪枝可降低视觉语言模型（VLM）的推理开销，但多数方法仅关注保留哪些标记。这种保留视角可能保留冗余高分标记，同时让被丢弃的证据缺乏近似代表。我们提出CoverPruner，一种无需训练的剪枝方法，从需求侧反向思考：一个标记被移除后，哪个现存原标记能代表它？CoverPruner将剪枝建模为表示覆盖最大化（RCM），在投影空间中以查询加权方式覆盖全部原始视觉标记集。其通过投影空间覆盖与轻量级第一层注意力探针实现RCM。在多个VLM架构和压缩率下，CoverPruner在所有对比方法中取得最佳平均准确率，且在激进压缩条件下提升最为明显。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。