Learning Deep Learning & Mini-Hackathon

In this session we look at building practical applications of deep learning. We first take a hands-on approach to learning about neural networks and how they function. Then we build on that knowledge by putting it into practice, building working applications. Finishing the session by allowing you to put your knowledge to the test for fun and prizes.

Session Program
Introduction to Deep Learning (1 hour)

• The Perceptron Game
• Gradient descent, backpropagation and optimization
• Deep learning hyperparameters

Applications of Deep Learning (1 hour)

• Regression analysis on tabular data
• Image analysis with convolutional neural networks
• Transfer learning for image classification (intermediate level)

Mini-Hackathon

• 3 Challenges - users may enter as individuals or teams (max 3) in 1 to 3 challenges.
  Challenge A) Regression analysis on tabular data. TEST - find the best training/test convergence for a regression problem on tabular data.
  Challenge B) Image analysis with CNN. TEST - Classify image dataset using CNN networks.
  Challenge C) Transfer learning for image classification. TEST - retrain an existing deep learning image classification model like AlexNet on a new image classification problem.

--- All three challenges will require hyperparameter optimization and possibly slight modifications to model architectures.
(a computing device is required for the challenge, all code will be in Python (Keras/TF2.0) and run on Google Colab, so no setup is required)

Prizes will be awarded to the top 3 contenders in each category.