Task 001/002 - Neuro-inspired Algorithms and Theory
|Event Date:||September 16, 2021|
Nitin Rathi, Purdue University Low Latency Spiking Neural Networks with Leakage and Threshold Optimization and its Applications in Sequential Learning Tasks
Abstract: Spiking Neural Networks (SNNs) present several advantages like low power computing with sparse spatio-temporal binary spikes, bio-inspired neuron model with an internal state (membrane potential), asynchronous operation, and low energy machine intelligence. However, these benefits are outweighed by the high inference latency. The increased latency results from inefficient input encoding and sub-optimal settings of neuron parameters (firing threshold and membrane leak). We propose DIET-SNN, a low-latency deep spiking network that is trained with gradient descent to optimize the membrane leak and the firing threshold along with other network parameters (weights). The membrane leak and threshold for each layer of the SNN are optimized with end-to-end backpropagation to achieve competitive accuracy at reduced latency. The analog pixel values of an image are directly applied to the input layer of DIET-SNN without the need to convert to spike-train. The first convolutional layer is trained to convert inputs into spikes where leaky-integrate-and-fire (LIF) neurons integrate the weighted inputs and generate an output spike when the membrane potential crosses the trained firing threshold. The trained membrane leak controls the flow of input information and attenuates irrelevant inputs to increase the activation sparsity in the convolutional and dense layers of the network. The reduced latency combined with high activation sparsity provides large improvements in computational efficiency. We evaluate DIET-SNN on image classification tasks from CIFAR and ImageNet datasets on VGG and ResNet architectures. We achieve top-1 accuracy of 69% with 5 timesteps (inference latency) on the ImageNet dataset with 12× less compute energy than an equivalent standard ANN.
In this talk, we also discuss the application of SNNs in sequential learning tasks on data from dynamic vision sensors (DVS) and natural language processing (NLP). In general sequential data is processed with complex RNNs (LSTM/GRU) with explicit feedback connections and an internal state to retain the previously seen data. Whereas neuron models in SNNs - integrate-and-fire (IF) or leaky-integrate-and-fire (LIF) - have implicit feedback in their internal state (membrane potential) by design and can be leveraged for sequential tasks. SNNs have 8× lower number of parameters than LSTMs resulting in smaller models and faster inference on memory-constrained devices. We evaluate SNN on gesture recognition from the IBM DVS dataset and sentiment analysis from the IMDB movie reviews dataset.
Nitin Rathi received a B.Tech. degree in electronics and communications engineering from the West Bengal University of Technology, Kolkata, India, in 2013. He is currently pursuing a Ph.D. degree in electrical and computer engineering at Purdue University, West Lafayette, IN, USA. He worked as a machine learning intern with GlobalFoundries, Santa Clara, during summer 2020 where he developed machine learning algorithms to identify defects from chip design images. His research interests include machine learning algorithms, neuromorphic computing, and energy-efficient deep learning.