wiki:2015/spikednn15

Spiking Implementations of Deep Neural Networks

Siegert-based LIF Training

Code: https://github.com/dannyneil/edbn

Paper: DBNs, Multimodal Fusion, Training Deep LIF Networks

Run example.m to train a sample small neural network with LIF neurons and watch it spike, classify, and generate.

High-Accuracy Conversion from ANN to SNN

Code: https://github.com/dannyneil/spiking_relu_conversion

Paper: High Accuracy Conversion with ReLUs

Train a typical NN to 98% accuracy, then convert it into a spiking neuron and watch how fast it can classify. Really recent work, and currently has the state-of-the-art accuracy for spiking implementations.

Relevant papers

  • O'Connor et al. 2013, Frontiers in Neuromorphic Engineering: First paper about spiking Deep Belief Networks, using Siegert method (Link)
  • Diehl et al. 2015, IJCNN: Recent paper about converting ANNs (ConvNets? + fully connected) to spiking NNs (Link)
  • Stromatias et al. 2015, IJCNN: Implementation of spiking DBNs on SpiNNaker
  • Neil and Liu, 2014, IEEE Trans. VLSI: FPGA implementation of spiking DBNs (Link)
  • Neftci et al. 2014, Frontiers in Neuromorphic Engineering: Event-based Contrastive Divergence learning rule (Link)
Last modified 4 years ago Last modified on 05/05/15 18:17:16

Attachments (2)

Download all attachments as: .zip