SparseCNN

A Tensorflow implementation of the How Can We Be So Dense? The Benefits of Using Highly Sparse Representations.

Implementation

This code was tested using TF2.0 and python 3.6.

pip install -r requirements.txt

This code only implement the simple CNN architectures SparseCNN1 and DenseCNN1

To test other architectures, you can modify the sparse_net.py or dense_net.py

To launch training SparseCNN1 as described in the paper

python train_evaluate.py

Results

As described in the paper, the sparseCNN shows more robustness to white noise in the inputs.