Knowledge Commons of Institute of Automation,CAS
Soft Threshold Ternary Networks | |
Xu, Weixiang1,2![]() ![]() ![]() ![]() ![]() ![]() | |
2020 | |
会议名称 | International Joint Conference on Artificial Intelligence |
会议日期 | 2020 |
会议地点 | Online |
摘要 | Large neural networks are difficult to deploy on mobile devices because of intensive computation and storage. To alleviate it, we study ternarization, a balance between efficiency and accuracy that quantizes both weights and activations into ternary values. In previous ternarized neural networks, a hard threshold∆ is introduced to determine quantization intervals. Although the selection of∆ greatly affects the training results, previous works estimate∆ via an approximation or treat it as a hyper-parameter, which is suboptimal. In this paper, we present the Soft Threshold Ternary Networks (STTN), which enables the model to automatically determine quantization intervals instead of depending on a hard threshold. Concretely, we replace the original ternary kernel with the addition of two binary kernels at training time, where ternary values are determined by the combination of two corresponding binary values. At inference time, we add up the two binary kernels to obtain a single ternary kernel. Our method dramatically outperforms current state-of-the-arts, lowering the performance gap between full-precision networks and extreme low bit networks. Experiments on ImageNet with AlexNet (Top-1 55.6%), ResNet-18 (Top-1 66.2%) achieves new state-of-the-art. |
收录类别 | EI |
语种 | 英语 |
七大方向——子方向分类 | AI芯片与智能计算 |
国重实验室规划方向分类 | 自主可控先进计算 |
文献类型 | 会议论文 |
条目标识符 | http://ir.ia.ac.cn/handle/173211/40624 |
专题 | 复杂系统认知与决策实验室_高效智能计算与学习 中国科学院自动化研究所 |
作者单位 | 1.Institute of Automation, Chinese Academy of Sciences 2.University of Chinese Academy of Sciences |
第一作者单位 | 中国科学院自动化研究所 |
推荐引用方式 GB/T 7714 | Xu, Weixiang,He, Xiangyu,Zhao, Tianli,et al. Soft Threshold Ternary Networks[C],2020. |
条目包含的文件 | 下载所有文件 | |||||
文件名称/大小 | 文献类型 | 版本类型 | 开放类型 | 使用许可 | ||
0318.pdf(387KB) | 会议论文 | 开放获取 | CC BY-NC-SA | 浏览 下载 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论