traNNsformers

TraNNsformer is an integrated MATLAB framework for training MLP and CNN networks using structured pruning approach to enable efficient mapping on memristive crossbar based neuromorphic architecures.

System requirements

Network Topologies

TraNNsformer framework is implemented using MATLAB to transform neural networks of two topologies:

1. Multi-layer perceptrons (MLP) models are specified and transformed using codes in traNNsformers/NN

2. Convolutional Neural networks (CNN) models are specified and transformed using codes in:

   1. if GPU support is not available and for small sized networks traNNsformers/CNN
   2. if GPU support is available and for large sized networks traNNsformers/CNN_wGPU

Citation

Please cite the following paper if you find this work useful:

• A. Ankit, T. Ibrayev, A. Sengupta, K. Roy. TraNNsformer: Clustered Pruning on Crossbar-based Architectures for Energy Efficient Neural Networks. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2019.

Authors

Aayush Ankit, Timur Ibrayev