Rate Coding Or Direct Coding: Which One Is Better For Accurate, Robust, And Energy-Efficient Spiking Neural Networks?

Kim, Youngeun; Park, Hyoungseob; Moitra, Abhishek; Bhattacharjee, Abhiroop; Venkatesha, Yeshwanth; Panda, Priyadarshini

doi:10.1109/icassp43922.2022.9747906

Cited by 33 publications

(24 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The PGD method only requires the sign of the gradient to generate adversarial perturbation, which is relatively insensitive to the exact value of the surrogate gradient in SNNs. Therefore, gradient-based methods for evaluating adversarial robustness have been adopted by the SNN community (Sharmin et al, 2020 ; Kundu et al, 2021 ; Kim et al, 2022 ; Nomura et al, 2022 ).…”

Section: Resultsmentioning

confidence: 99%

Enhancing spiking neural networks with hybrid top-down attention

Liu

Zhao

2022

Front. Neurosci.

View full text Add to dashboard Cite

As the representatives of brain-inspired models at the neuronal level, spiking neural networks (SNNs) have shown great promise in processing spatiotemporal information with intrinsic temporal dynamics. SNNs are expected to further improve their robustness and computing efficiency by introducing top-down attention at the architectural level, which is crucial for the human brain to support advanced intelligence. However, this attempt encounters difficulties in optimizing the attention in SNNs largely due to the lack of annotations. Here, we develop a hybrid network model with a top-down attention mechanism (HTDA) by incorporating an artificial neural network (ANN) to generate attention maps based on the features extracted by a feedforward SNN. The attention map is then used to modulate the encoding layer of the SNN so that it focuses on the most informative sensory input. To facilitate direct learning of attention maps and avoid labor-intensive annotations, we propose a general principle and a corresponding weakly-supervised objective, which promotes the HTDA model to utilize an integral and small subset of the input to give accurate predictions. On this basis, the ANN and the SNN can be jointly optimized by surrogate gradient descent in an end-to-end manner. We comprehensively evaluated the HTDA model on object recognition tasks, which demonstrates strong robustness to adversarial noise, high computing efficiency, and good interpretability. On the widely-adopted CIFAR-10, CIFAR-100, and MNIST benchmarks, the HTDA model reduces firing rates by up to 50% and improves adversarial robustness by up to 10% with comparable or better accuracy compared with the state-of-the-art SNNs. The HTDA model is also verified on dynamic neuromorphic datasets and achieves consistent improvements. This study provides a new way to boost the performance of SNNs by employing a hybrid top-down attention mechanism.

show abstract

Section: Resultsmentioning

confidence: 99%

Enhancing spiking neural networks with hybrid top-down attention

Liu

Zhao

2022

Front. Neurosci.

View full text Add to dashboard Cite

show abstract

“…There are several methods for encoding data into spikes. For example, Kim et al [42] showed that using a fully-connected method to encode data into spikes yields better results for the SNN training than the popular rate encoding, discussed in the next subsection. Another important realization from their work is the impact of different encoding methods on the performance of the SNNs.…”

Section: Spike Encodingmentioning

confidence: 99%

“…The first model we compare is proposed by Youngeun et al [42], discussing direct encoding. In their paper the authors show that direct encoding is more accurate than rate encoding.…”

Section: Classificationmentioning

confidence: 99%

Spiking Neural Operators for Scientific Machine Learning

Kahana¹,

Zhang²,

Leonard³

et al. 2022

Preprint

View full text Add to dashboard Cite

One of the main broad applications of deep learning is function regression. However, despite their demonstrated accuracy and robustness, modern neural network architectures require heavy computational resources to train. One method to mitigate or even resolve this inefficiency has been to draw further inspiration from the brain and reformulate the learning process in a more biologicallyplausible way, developing what are known as Spiking Neural Networks (SNNs), which have been gaining traction in recent years. In this paper we present an SNN-based method to perform regression, which has been a challenge due to the inherent difficulty in representing a function's input domain and continuous output values as spikes. We use a DeepONet -neural network designed to learn operators -to learn the behavior of spikes. Then, we use this approach to do function regression. We propose several methods to use a DeepONet in the spiking framework, and present accuracy and training time for different benchmarks.

show abstract

“…3(b) shows the details of the direct coding. A recent study [48] systematically analyzed these two coding schemes in terms of accuracy, robustness, and energy efficiency. Through experiments, it was found that direct coding can achieve higher classification accuracy over shorter time steps.…”

Section: Spiking Encoder Schemesmentioning

confidence: 99%

“…We test the classification performance of the two submodules on six HSI data sets. The experimental parameters' spatial size is set to 13, the time steps are set to 30, and the experiments are performed The second is that the encoding method of our model adopts direct encoding, and the direct encoding method will be affected by noise to a certain extent [48]. However, overall, the model still has certain advantages.…”

Section: E Model Evaluationmentioning

confidence: 99%

Hyperspectral Image Classification of Brain-Inspired Spiking Neural Network Based on Approximate Derivative Algorithm

Liu

Cao

et al. 2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Recently, deep learning methods have made significant progress in solving hyperspectral images (HSIs) classification problems of high-dimensional features, band redundancy, and spectral mixture. However, the deep neural network is too complex, with a long training time and high energy consumption, making it difficult to deploy on edge computing devices. In order to solve the above problems, this paper proposes a brain-inspired computing framework based on the spiking leaky integrate-andfire neuron model for HSIs classification. Then we design an approximate derivative algorithm to solve the non-differentiable spike activity of the spiking neuron. The framework uses direct coding to generate spatiotemporal spikes for input HSI and achieves efficient extraction of spatial-spectral features through spiking standard convolution and spiking depthwise separable convolution. Extensive experiments are performed on four benchmark hyperspectral data sets and two public unmanned aerial vehicle-borne hyperspectral data sets. Experiments show that the proposed model has the advantages of high classification accuracy and fewer spiking time steps. The proposed model can save about 10 times computational energy consumption compared with the CNN of the same architecture. This research has great significance for overcoming the technical bottleneck of HSI classification based on brain-inspired computing, solving the critical problems of mobile computing in unmanned autonomous systems, and realizing the engineering application of unmanned aerial vehicles and software-defined satellites. The source code will be made available at https://github.com/Katherine-Cao/HSI SNN.

show abstract

Rate Coding Or Direct Coding: Which One Is Better For Accurate, Robust, And Energy-Efficient Spiking Neural Networks?

Cited by 33 publications

References 17 publications

Enhancing spiking neural networks with hybrid top-down attention

Enhancing spiking neural networks with hybrid top-down attention

Spiking Neural Operators for Scientific Machine Learning

Hyperspectral Image Classification of Brain-Inspired Spiking Neural Network Based on Approximate Derivative Algorithm

Contact Info

Product

Resources

About