Sparse Accelerator

Sparse accelerators are hardware designed to efficiently process sparse neural networks, aiming to reduce computational cost and energy consumption while maintaining accuracy. Current research focuses on optimizing sparse matrix multiplication for various neural network architectures, including vision transformers, spiking neural networks, and convolutional neural networks, employing techniques like N:M sparsity and weight pruning. These advancements are significant for deploying deep learning models on resource-constrained devices and improving the efficiency of large-scale training and inference tasks in diverse applications such as image recognition and natural language processing. The resulting speedups and energy savings are substantial, demonstrating the practical impact of this research area.

Papers

September 15, 2024

ELSA: Exploiting Layer-wise N:M Sparsity for Vision Transformer Acceleration
Ning-Chi Huang, Chi-Chih Chang, Wei-Cheng Lin, Endri Taka, Diana Marculescu, Kai-Chiang Wu
Layer Wise Sparse Matrix Multiplication Sparsity Level Sparsity Aware Sparse Accelerator

July 19, 2024

LoAS: Fully Temporal-Parallel Dataflow for Dual-Sparse Spiking Neural Networks
Ruokai Yin, Youngeun Kim, Di Wu, Priyadarshini Panda
Spiking Neural Network Parallel Data DNN Workload Sparse Accelerator

September 22, 2023

Efficient N:M Sparse DNN Training Using Algorithm, Architecture, and Dataflow Co-Design
Chao Fang, Wei Sun, Aojun Zhou, Zhongfeng Wang
Practical Algorithm Architecture Design Better Zero Sparse Training Dataflow Architecture Sparse Accelerator Sparse DNN

June 28, 2023

An Efficient Sparse Inference Software Accelerator for Transformer-based Language Models on CPUs
Haihao Shen, Hengyu Meng, Bo Dong, Zhe Wang, Ofir Zafrir, Yi Ding, Yu Luo, Hanwen Chang, Qun Gao, Ziheng Wang, Guy Boudoukh, Moshe Wasserblat
Transformer Based Language Model Neural Network Inference Sparse Accelerator Sparse Learning Method

February 28, 2023

AccelTran: A Sparsity-Aware Accelerator for Dynamic Inference with Transformers
Shikhar Tuli, Niraj K. Jha
Transformer Megatron Decepticons Transformer Model Transformer Inference Dynamic Inference Accelerator Architecture Transformer Accelerator Sparse Accelerator

October 29, 2022

LearningGroup: A Real-Time Sparse Training on FPGA via Learnable Weight Grouping for Multi-Agent Reinforcement Learning
Je Yang, JaeUk Kim, Joo-Young Kim
Multi Agent Reinforcement Learning Field Programmable Gate Array Single Agent Reinforcement Learning Weight Freezing Dynamic Sparse Training Group Learning Better on Chip Data Sparse Accelerator

April 21, 2022

TorchSparse: Efficient Point Cloud Inference Engine
Haotian Tang, Zhijian Liu, Xiuyu Li, Yujun Lin, Song Han
Point Cloud 3D Point Cloud Sparse Convolution Sparse Accelerator

November 9, 2021

Phantom: A High-Performance Computational Core for Sparse Convolutional Neural Networks
Mahmood Azhar Qureshi, Arslan Munir
Phantom 2D Accelerator Phantom Threat Sparse Convolutional Sparse Accelerator Sparse CNN