CSE 176 Project 2 - CNN Compression using LC Algorithm toolkit

Project Details

• Dataset: MNIST.mat
• Model: LeNet5-Caffe convolutional neural network
• Tasks
  • Multiclass classification over 5 digits: 1, 2, 4, 7, 8
  • Finding optimal hyperparamters over SGD (batch size, learning rate, momentum, L2 regulatization
  • Finding optimal hyperparamters over LC algorithm (# of L steps, # of LC steps, mu-schedule)
• Sources:
  • Prof. MACP's datasets on his faculty website
  • LC Model Compression toolkit repository and website
  • Image source

Interactive Demo

1. Visit the following link: Google Colab
2. Run the first two codeblocks before the "IMPORTANT!" heading.
3. Go to "Runtime>Restart Runtime" and click Yes.
4. Click and run each codeblock in order.