H
Hong Ma
Researcher at Chinese Academy of Sciences
Publications - 7
Citations - 127
Hong Ma is an academic researcher from Chinese Academy of Sciences. The author has contributed to research in topics: Convolutional neural network & Spiking neural network. The author has an hindex of 4, co-authored 7 publications receiving 76 citations.
Papers
More filters
Proceedings ArticleDOI
FBNA: A Fully Binarized Neural Network Accelerator
TL;DR: This work proposes the first fully binarized Convolutional neural network accelerator (FBNA) architecture, in which all convolutional operations are binarization and unified, even including the first layer and padding, which provides more resource, parallelism and scalability optimization opportunities.
Proceedings ArticleDOI
MaPU: A novel mathematical computing architecture
Donglin Wang,Xueliang Du,Leizu Yin,Chen Lin,Hong Ma,Weili Ren,Wang Huijuan,Xingang Wang,Shaolin Xie,Wang Lei,Zijun Liu,Tao Wang,Zhonghua Pu,Guangxin Ding,Zhu Mengchen,Lipeng Yang,Ruoshan Guo,Zhiwei Zhang,Xiao Lin,Jie Hao,Yang Yongyong,Wenqin Sun,Fabiao Zhou,NuoZhou Xiao,Qian Cui,Xiaoqin Wang +25 more
TL;DR: The MaPU architecture is presented, a novel architecture which is suitable for data-intensive computing with great power efficiency and sustained computation throughput, and increases the actual power efficiency by an order of magnitude comparable with the traditional CPU and GPGPU.
Proceedings ArticleDOI
Fast and Efficient Deep Sparse Multi-Strength Spiking Neural Networks with Dynamic Pruning
TL;DR: An innovative deep multi-strength SNN (M-SNN) structure is proposed which relaxes the restriction of the neuron output spike strength while the event-driven feature for low-power implementations is maintained and can be converted from CNN with comparable accuracy and fast inference speed.
Journal ArticleDOI
A High-Efficiency FPGA-Based Accelerator for Binarized Neural Network
TL;DR: The convolutional neural network has exhibited outstanding performance in various applications, but the deployment of CNN on embedded and mobile devices is limited by the massive computations required.
Proceedings ArticleDOI
Low Latency Spiking ConvNets with Restricted Output Training and False Spike Inhibition
TL;DR: This paper proposes a restricted output training method to normalize the converted weights dynamically in the CNN-SNN training phase and proposes a temporal max pooling method to approximate the max Pooling operation in ConvNets without accuracy loss.