用向量量化压缩KAN模型，无需重训练即可大幅减少边缘设备内存占用。
SHARe-KAN: Post-Training Vector Quantization for Cache-Resident KAN Inference
- 将KAN的样条系数分解为增益-形状-偏置，共享码本实现高效压缩。
- 在PASCAL VOC上压缩140倍，存储仅6.32MB，精度损失2.0点（80.22% mAP）。
- 适合部署多任务边缘AI模型，支持零样本迁移和低精度量化。
预训练视觉Kolmogorov-Arnold网络（KAN）在每条边存储密集的B样条网格，使预测头参数量相比同规模MLP增加140倍以上，导致边缘加速器推理陷入内存瓶颈。标准幅值剪枝在此类模型上失效：零样本稀疏化会严重损失精度，恢复需迭代微调，不适用于部署场景。本文提出SHARe-KAN，一种后训练编译器，通过增益-形状-偏置分解与层共享码本压缩样条系数，并结合LUTHAM运行时将码本映射至片上L2缓存。在基于ResNet-50主干的PASCAL VOC检测任务中，SHARe-KAN Int8相比密集型KAN基线实现9.3倍存储压缩（6.32 MB vs. 58.67 MB预测头），精度下降2.0点（mAP从82.22%降至80.22%），且无需重新训练。零样本迁移至COCO仍保留88.9%的原模型精度；大部分损失来自向量量化聚类步骤本身，而从FP32到Int8的进一步量化仅导致1.3点精度下降。该方法在规模扩大时优势更显著：当拥有50个任务头时，密集型KAN预测头存储达2.9 GB，而SHARe-KAN Int8仅需211 MB，实现13.9倍压缩，使多专家KAN在现有边缘芯片内存预算内可部署。
原文摘要 · Abstract (English)
Pre-trained Vision Kolmogorov-Arnold Networks (KANs) store a dense B-spline grid on every edge, inflating prediction-head parameter counts by more than 140X relative to a comparable MLP and pushing inference into a memory-bound regime on edge accelerators. Standard magnitude pruning fails on these pre-trained models: zero-shot sparsity collapses accuracy, and restoring it requires an iterative fine-tuning loop that is impractical in deployment settings. We present SHARe-KAN, a post-training compiler that compresses spline coefficients via a Gain-Shape-Bias decomposition with a layer-shared codebook, paired with LUTHAM, an ExecuTorch runtime that maps the codebook into on-chip L2. On PASCAL VOC detection with a ResNet-50 backbone, SHARe-KAN Int8 reaches 9.3X storage compression over the Dense KAN baseline (6.32 MB vs. 58.67 MB prediction head) at a 2.0 point in-domain accuracy cost (80.22% vs. 82.22% mAP), with no retraining. Zero-shot transfer to COCO retains 88.9% of the Dense KAN mAP; most of this gap comes from the VQ clustering step itself, and further quantization from FP32 to Int8 costs only 1.3 retention points. The value of the approach compounds at scale: at 50 task heads, Dense KAN prediction-head storage reaches 2.9 GB while SHARe-KAN Int8 requires 211 MB, a 13.9X reduction that brings multi-expert KAN deployment within the memory budgets of contemporary edge silicon.
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。