Going Deeper With Directly-Trained Larger Spiking Neural Networks
Authors: Hanle Zheng, Yujie Wu, Lei Deng, Yifan Hu, Guoqi Li11062-11070
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we report superior accuracy results including 93.15% on CIFAR-10, 67.8% on DVS-CIFAR10, and 67.05% on Image Net with very few timesteps. |
| Researcher Affiliation | Academia | Hanle Zheng 1, Yujie Wu 1, Lei Deng 1,3, Yifan Hu 1 and Guoqi Li 1,2 1 Center for Brain-Inspired Computing Research, Department of Precision Instrument, Tsinghua University, Beijing 100084, China 2 Beijing Innovation Center for Future Chip, Tsinghua University, Beijing 100084, China 3 Department of Electrical and Computer Engineering, University of California, Santa Barbara |
| Pseudocode | No | No pseudocode or algorithm blocks are formally labeled or presented within the main body of the paper. The paper mentions that 'The codes of the td BN can be found in Supplementary Material A.' and 'The codes of the overall training algorithm are shown in Supplementary Material A.', referring to actual code, not pseudocode in the main text. |
| Open Source Code | Yes | The codes of the td BN can be found in Supplementary Material A. The codes of the overall training algorithm are shown in Supplementary Material A. |
| Open Datasets | Yes | CIFAR-10 is an image dataset with 50000 training images and 10000 testing images with size of 32 32, which all belong to 10 classes. ... Image Net (Deng et al. 2009) contains 1.28 million training images and 50000 validating images. ... DVS-Gesture (Amir et al. 2017) is a collection of moving gestures ... DVS-CIFAR10 (Li et al. 2017) is a neuromorphic dataset converted from famous CIFAR-10 to its dynamic form. |
| Dataset Splits | Yes | Image Net (Deng et al. 2009) contains 1.28 million training images and 50000 validating images. |
| Hardware Specification | No | No specific hardware details (like GPU/CPU models, memory, or cloud instance types) used for running experiments are mentioned in the paper. |
| Software Dependencies | No | The paper mentions 'ANN-oriented programming frameworks (e.g. Pytorch)' but does not specify any software names with version numbers. It refers to supplementary material for code, but the main text lacks versioned dependencies. |
| Experiment Setup | Yes | In this experiment, we use Res Net-19 with timesteps of 2, 4, 6. ... On Image Net, we test Res Net-34 with standard size and large size. The large model doubles the channels compared with the standard one and achieves 67.05% top-1 accuracy with just 6 timesteps. Also, we use Res Net-50 to explore the very deep directly-trained SNNs and achieve 64.88% top-1 accuracy. ... We set timestep T to be 40. |