.png){kind=icon}
.png){kind=icon}
Return
Error-Aware Conversion from ANN to SNN via Post-training Parameter Calibration
.png){kind=icon}
PRE
AI
DOI：10.1007/s11263-024-02046-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Spiking Neural Network (SNN), originating from the neural behavior in biology, has been recognized as one of the next-generation neural networks. Conventionally, SNNs can be obtained by converting from pre-trained Artificial Neural Networks (ANNs) by replacing the non-linear activation with spiking neurons without changing the parameters. In this work, we argue that simply copying and pasting the weights of ANN to SNN inevitably results in activation mismatch, especially for ANNs that are trained with batch normalization (BN) layers. To tackle the activation mismatch issue, we first provide a theoretical analysis by decomposing local layer-wise conversion error, and then quantitatively measure how this error propagates throughout the layers using the second-order analysis. Motivated by the theoretical results, we propose a set of layer-wise parameter calibration algorithms, which adjusts the parameters to minimize the activation mismatch. To further remove the dependency on data, we propose a privacy-preserving conversion regime by distilling synthetic data from source ANN and using it to calibrate the SNN. Extensive experiments for the proposed algorithms are performed on modern architectures and large-scale tasks including ImageNet classification and MS COCO detection. We demonstrate that our method can handle the SNN conversion and effectively preserve high accuracy even in 32 time steps. For example, our calibration algorithms can increase up to 63% accuracy when converting MobileNet against baselines.
Keywords:
Brain-inspired computing
Spiking Neural Networks
ANN-SNN conversion
Hessian-based perturbation analysis
Journal
IF:
9.3
Papers:
3.9K
Citations:
2.8W