用可调误差的软稀疏技术，大幅降低CNN计算量且不损失精度。
Hardware Efficient Approximate Convolution with Tunable Error Tolerance for CNNs
- 用最高有效位代理跳过微小非零乘法，实现硬件友好的软稀疏。
- 在LeNet-5上将ReLU乘加操作减少88.42%，Tanh减少74.87%无精度损失。
- 适合资源受限设备部署，特别适用于边缘端低功耗推理场景。
现代CNN的高计算需求阻碍了其在边缘设备上的部署。传统‘硬稀疏’（跳过数学零）在深层网络或平滑激活函数（如Tanh）下效果下降。本文提出一种‘软稀疏’范式，利用硬件高效的最高有效位（MSB）代理跳过可忽略的非零乘法。该方法作为自定义RISC-V指令集成，在LeNet-5（MNIST）上实现了ReLU MACs减少88.42%，Tanh MACs减少74.87%，且零精度损失——性能优于零跳过方法5倍。通过门控未激活乘法器，估计在ReLU和Tanh下分别节省35.2%和29.96%功耗。尽管内存访问导致功耗降低非线性，但该方法显著优化了资源受限环境下的推理效率。
原文摘要 · Abstract (English)
Modern CNNs' high computational demands hinder edge deployment, as traditional ``hard'' sparsity (skipping mathematical zeros) loses effectiveness in deep layers or with smooth activations like Tanh. We propose a ``soft sparsity'' paradigm using a hardware efficient Most Significant Bit (MSB) proxy to skip negligible non-zero multiplications. Integrated as a custom RISC-V instruction and evaluated on LeNet-5 (MNIST), this method reduces ReLU MACs by 88.42% and Tanh MACs by 74.87% with zero accuracy loss--outperforming zero-skipping by 5x. By clock-gating inactive multipliers, we estimate power savings of 35.2% for ReLU and 29.96% for Tanh. While memory access makes power reduction sub-linear to operation savings, this approach significantly optimizes resource-constrained inference.
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。