Intel INRC workshop

ANN-SNN conversion framework on Loihi

Gregor Lenz, Sio-Hoi Ieng, Ryad Benosman

• lighweight wrapper for nxSDK
   
• modelled after pyTorch
   
• interactive flow
   
• supports batching, batchnorm folding

Conversion framework

Layer types

• Dense, MaxPooling, Convolutional: kernel size, strides, padding, groups, skip connections

Demo

github.com/biphasic

@gregorlenz

gregor.lenz@inserm.fr