TBCAS-Towards-Healthcare-and-Biomedical-Applications

Supplementary Repository for Hardware Implementation of Deep Network Accelerators Towards Healthcare and Biomedical Applications, which has been submitted to IEEE Transactions on Biomedical Circuits and Systems (TBioCAS).

1. Baseline.ipynb reproduces the baseline implementations from https://github.com/Enny1991/dvs_emg_fusion/ using PyTorch instead of Keras/TensorFlow.
2. Novel Simulations.ipynb reproduces results presented in Section III, where MemTorch is used to similate Memristive Deep Neural Networks (MDNNs).
3. Files in FPGA can be used to reproduce results presented in Section III, where PipeCNN is used to deploy trained networks on FPGA. The compilation flow used is as follows:
   1. Trained parameters of each network, state_dict, are exported to serialized pt objects on disk using PyTorch.
   2. Serialized pt objects are converted to equivalent protxt and caffemodel files using PytorchToCaffe.
   3. Fixed-Point caffe models are generated using MATLAB's Fixed-point toolbox, and stored as single binary dat files, as described here.
   4. The dat files are deployed using PipeCNN, as described here.