Event-Clock Hybrid Driven Reconfigurable Perception-Computation Technology

事件时钟混合驱动的可重构感知计算技术

• 批准号: 22K21280
• 负责人: KAN YIRONG
• 金额: $ 1.83万
• 依托单位: Nara Institute of Science and Technology
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Research Activity Start-up
• 财政年份: 2022
• 资助国家: 日本
• 起止时间: 2022-08-31 至 2024-03-31
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-22K21280/
• 关键词: Reconfigurable Computing; CGRA; Neuromorphic Systems; Spiking Neural Networks; Hybrid Driven

This year, we developed and verified the following technologies for the hybrid-driven reconfigurable perception-computation platform: (1) Spike coding of Electroencephalogram (EEG) signals and its spiking neural network (SNN)-based processing. In several works, we successfully applied spike coding to adaptive, stochastic and frequency coding of EEG signals, respectively, and achieved competitive sleep stage classification accuracy based on SNN; (2) A ternary weight quantization method for deep SNNs and hardware implementation. In this work, we achieved energy-efficient inference hardware by quantizing the weights of SNNs to {-1, 0, 1}. The gradient disappearance problem during model training is avoided by designing cross-layer connections. Simple logical operations can be used in ternary weights SNNs at the inference stage, to reducing hardware overhead; (3) Training and construction mechanism of reconfigurable bisection neural network (BNN) topology. We proposed a general construction method of BNN and its training mechanism. By constructing a mask matrix with a bisection structure, we can automatically train a BNN model with a specific topology.

今年，我们为混合驱动的可重构感知计算平台开发并验证了以下技术：(1)脑电信号的棘波编码及其基于脉冲神经网络(SNN)的处理。在几项工作中，我们成功地将尖峰编码分别应用于脑电信号的自适应编码、随机编码和频率编码，获得了基于SNN的具有竞争力的睡眠阶段分类精度；(2)一种适用于深度SNN的三值加权量化方法及其硬件实现。在这项工作中，我们通过将SNN的权重量化到{-1，0，1}来实现节能的推理硬件。通过设计跨层连接，避免了模型训练过程中的梯度消失问题。在推理阶段，可以在三值权值神经网络中使用简单的逻辑运算，以减少硬件开销；(3)可重构二分神经网络(BNN)拓扑的训练和构造机制。提出了一种神经网络的一般构造方法及其训练机制。通过构造具有二等分结构的掩码矩阵，可以自动训练具有特定拓扑结构的BNN模型。

项目成果

Bisection Neural Network Toward Reconfigurable Hardware Implementation

• DOI: 10.1109/tnnls.2022.3195821
• 发表时间: 2022-08
• 期刊: IEEE Transactions on Neural Networks and Learning Systems
• 影响因子: 10.4
• 作者: Yan Chen;Renyuan Zhang;Yirong Kan;Sa Yang;Y. Nakashima

Adaptive spike-like representation of eeg signals for sleep stages scoring

用于睡眠阶段评分的脑电图信号的自适应尖峰状表示

• 发表时间: 2022
• 作者: Lingwei Zhu;Ziwei Yang;Koki Odani;Guang Shi;Yirong Kan;Zheng Chen;Renyuan Zhang
• 通讯作者: Renyuan Zhang

MuGRA: A Scalable Multi-Grained Reconfigurable Accelerator Powered by Elastic Neural Network

MuGRA：由弹性神经网络提供支持的可扩展多粒度可重构加速器

• DOI: 10.1109/tcsi.2021.3099034
• 发表时间: 2022
• 期刊: IEEE Transactions on Circuits and Systems I: Regular Papers
• 作者: Kan Yirong;Wu Man;Zhang Renyuan;Nakashima Yasuhiko
• 通讯作者: Nakashima Yasuhiko

Automatic Sleep Staging via Frequency-Wise Spiking Neural Networks

通过频率尖峰神经网络自动睡眠分期

• 发表时间: 2022
• 作者: Haohui Jia;Ziwei Yang;Pei Gao;Man Wu;Chen Li;Yirong Kan;Renyuan Zhang
• 通讯作者: Renyuan Zhang

A Stochastic Coding Method of EEG Signals for Sleep Stage Classification

• DOI: 10.1109/socc56010.2022.9908121
• 发表时间: 2022-09
• 期刊: 2022 IEEE 35th International System-on-Chip Conference (SOCC)
• 作者: Guangxian Zhu;Huijian. Wang;Yirong Kan;Zheng Chen;Ming Huang;M. Altaf-Ul-Amin;N. Ono;Shigehiko Kanaya;Renyuan Zhang;Y. Nakashima