MNIST_ASIC_LAB2

MNIST classification using binary neural nets in pytorch. Made for LAB2 of Hardware AI of TUDelft

INSTALLATION PROCEDURE

1. Install Conda in your computer to manage the python environment from here.
2. Verify installation was succesful by issuing conda list in the terminal.
3. Create a new environment with conda create -n binary_net python=3.9, where binary_net is the name of your environment.
4. Activate enviroment with conda activate binary_net
5. Install pytorch and torchvision with with pip3 install torch torchvision.
6. Install juppyter notebook in your environment with conda install jupyter.
7. Install matplotlib with conda install -c conda-forge matplotlib.
8. You should now be able to use jupyter notebook without having issue with libraries.
9. Lunch the notebook with jupyter notebook.
10. We will work on the file called our_network.ipynb. `

Files to use for hardware testing

Files below are the weights for a 4 layer network with 2 hidden layer considered as baseline (you can find them in /baseline)

W_fc1.csv -> input layer (IN=784,OUT=100)

W_fc2.csv -> hidden1 layer (IN=100,OUT=100)

W_fc3.csv -> hidden2 layer (IN=100,OUT=100)

W_fc4.csv -> output layer (IN=100,OUT=10)

File below is the flatten(vectorized) input of the image 2

input_image_flatten.csv -> input image (ROW=1,COL=784)

With this files we should be able to validate that our machine can classify the image of number 2. This is verified in software.