This contains the course project codes for EE5111 at IIT-Madras
The code contains the linear version of the BCDL code for the paper:
Louizos, Christos, Karen Ullrich, and Max Welling. "Bayesian compression for deep learning." Advances in Neural Information Processing Systems. 2017.
Although the backbone of the code is the same as in the official repo, some aspects of the code have been modified
For getting pre-trained model, run:
python mnist_nn.py
This will save a pretrained neural network that will be used by the bayesian network (BN) for initialization.
For running the BN with random initialization, run:
python example.py
For running the same BN initialized with pretrained weights, run:
python example.py --load_pretrained
Additionally, you can prefix the above command with 'CUDA_VISIBLE_DEVICES=X' to select the Xth GPU on your system. For example, running the following command will train the BN initialized with pretrained weights on GPU 1:
CUDA_VISIBLE_DEVICES=1 python example.py --load_pretrained
To play around with the code and the trained models on CPU, use the notebook BCDL_playground.ipynb.