SATA: Sparsity-Aware Training Accelerator for Spiking Neural Networks

Yin, Ruokai; Moitra, Abhishek; Bhattacharjee, Abhiroop; Kim, Youngeun; Panda, Priyadarshini

doi:10.1109/tcad.2022.3213211

Cited by 20 publications

(15 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Finally, we introduce the hardware platform that we design for carrying out the experiments on energy efficiency. We extend the overall architecture and PE design from Yin et al ( 2022 ) to support the necessary computation and data movement for our SNNs in HAR tasks. Owing to the 1D convolution and temporal dynamics that are naturally embedded in the time series data, the complexity of the hardware design has been largely reduced.…”

Section: Methodsmentioning

confidence: 99%

“…The size of the PE array and global buffers are configurable according to different network structures. In this work, we set the number of PEs to 128, weight (W) buffer to 32 KB, and spike (S) buffer to 576 bytes, for matching with the dataflow used in Yin et al ( 2022 ). We briefly explain the computation and data movement flow below.…”

Section: Methodsmentioning

confidence: 99%

“…The latest membrane potential for time step t is then sent to the LIF unit at step ❺ to generate the output spike s ( t ) . According to the dataflow in Yin et al ( 2022 ), each PE will only focus on working on one output neuron, and the PE array processes the whole output feature map in parallel. After that, spikes for the next layer will be written into the S buffer at step ❻, and the whole process will repeat.…”

Section: Methodsmentioning

confidence: 99%

“…The second metric in hardware performance is energy consumption. We estimate the energy consumption by simulating the proposed hardware design in Section 2.5 together with our ReLU-based ANN baseline through the energy simulator proposed in Yin et al ( 2022 ). The overall energy we consider consists of two parts: computing energy and data-moving energy.…”

Section: Methodsmentioning

confidence: 99%

“…Additionally, we extend a previous hardware accelerator design to support 1D convolution along the time dimension, making it suitable for SNN implementation (Yin et al, 2022 ). We evaluate our SNNs on three widely-used HAR datasets (UCI-HAR Anguita et al, 2013 , UniMB SHAR Micucci et al, 2017 , HHAR Stisen et al, 2015 ) and compare them with ANN baselines.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Efficient human activity recognition with spatio-temporal spiking neural networks

Li,

Yin,

Kim

et al. 2023

Front. Neurosci.

Self Cite

View full text Add to dashboard Cite

In this study, we explore Human Activity Recognition (HAR), a task that aims to predict individuals' daily activities utilizing time series data obtained from wearable sensors for health-related applications. Although recent research has predominantly employed end-to-end Artificial Neural Networks (ANNs) for feature extraction and classification in HAR, these approaches impose a substantial computational load on wearable devices and exhibit limitations in temporal feature extraction due to their activation functions. To address these challenges, we propose the application of Spiking Neural Networks (SNNs), an architecture inspired by the characteristics of biological neurons, to HAR tasks. SNNs accumulate input activation as presynaptic potential charges and generate a binary spike upon surpassing a predetermined threshold. This unique property facilitates spatio-temporal feature extraction and confers the advantage of low-power computation attributable to binary spikes. We conduct rigorous experiments on three distinct HAR datasets using SNNs, demonstrating that our approach attains competitive or superior performance relative to ANNs, while concurrently reducing energy consumption by up to 94%.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%