Max-Pooling Operations in Deep Spiking Neural Networks
Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, Switzerland
What is THIS?
The attempts of converting Analog Neural Networks to Spiking Neural Networks started in 2012. And the first successful sign of this conversion is proposed in 2014. All previous papers tried to use Average Pooling over Max-Pooling because of its simplicity.
In this report, I proposed three ways of implementing Max-Pooling by adding a gating function that monitors the neurons’ activities.
The Spiking Max-Pooling operation is included in
snn_toolbox, clone the code from here:
$ git clone https://github.com/NeuromorphicProcessorProject/snn_toolbox
Special thanks to Bodo Rueckauer for releasing the project.
What’s the RESULT?
In our paper that appears in the Computing with Spikes NIPS 2016 Workshop, we reported the best SNN results to date (2016).
Paper in ANN-SNN Conversion
by Rueckauer, B., Lungu, I-A, Hu, Y., Pfeiffer, M.
Journal Paper in ANN-SNN Conversion
by Rueckauer B., Lungu, I-A., Hu, Y., Pfeiffer, M., Liu, S-C.
Hu, Y. (2016). Max-Pooling Operations in Deep Spiking Neural Networks. NSC Short Project Report. Zürich, Switzerland: University of Zürich and ETH Zürich.
Rueckauer, B., Lungu, I-A, Hu, Y., Pfeiffer, M. (2016). Theory and Tools for the Conversion of Analog to Spiking Convolutional Neural Networks. In Computing with Spikes NIPS 2016 Workshop, Barcelona, Spain, 2016.
Rueckauer B., Lungu, I-A., Hu, Y., Pfeiffer, M., Liu, S-C. (2017). Conversion of Continuous-Valued Deep Networks to Efficient Event-Driven Networks for Image Classification. Frontiers in Neuroscience, 11:682.