Experiments
Here we document our experiments, reproducing classic papers and testing new ideas.
Experiment Logs
LeNet-1 Reproducibility
reproducing the original LeNet-1 results on MNIST and comparing with the 1998 paper.
Reproducibility Studies
Multi-Channel CNN
Experimenting how the convolution operation works with multiple channels, then trying to add the channels first, from 3 channels to 1 channel, and then applying the convolution, and comparing the results to see which one performs better and why.
Building Models
[English - RNN Character Generation]
Simple expermint with chars generation using english text. See the notebook: View Training Notebook
Live Demo - RNN Character Generation
Interactive demo of a character-level RNN that generates text based on user input.
Arabic Character Generation
Version 1:
Character-level RNN trained on Arabic text to generate new Arabic sentences. View Training Notebook
Version 2:
Character-level RNN trained on Arabic text with gradient norm tracking to analyze potential vanishing gradient issues. View Training Notebook
LSTM for Arabic Text Generation
In this experiment, we train an LSTM to generate Arabic text character by character. The notebook contains a step by step walkthrough of the code, and what does each line do. You can find the notebook here
Comments