Saving the Princess with Deep Learning
Submitted by Navin (@navinpai) on Monday, 10 April 2017
Deep Learning has provided an entirely new paradigm of solving problems which were otherwise deemed difficult to solve and is widely seen as a strong leap towards AGI. The field is moving at a rapid pace and innovative solutions to problems keep coming up every day.
In this talk, we will look at how to use the paradigm to do something useful AND fun. We’ll teach a deep neural network to learn how to play Super Mario. As part of the talk, we’ll not only do a dive into the code, learn about different Python Deep Learning frameworks, and talk about training the system but also look at the optimisations done. And of course, we’ll do a demo of the system (and share code!)
- Q Learning and DQN, and why this is amazing!
- Frameworks we use
- Problem definition
- Breaking down the problem for our Neural Network
- Designing the network
- Training, Validation and Testing
- Hurdles, Solutions and Optimisations
Basic understanding of Linear Algebra + Neural Networks is preferable but not essential (As I’ll cover it anyway and provide links in slides for explanations)
Navin is a graduate of the International Institute of Information Technology - Bangalore (IIIT-B) where he specialised in Machine Learning and AI. He has been a Pythonista for over half a decade and has conducted introductory classes in Python for college students many times. He writes about his experiments (in code and otherwise) at LifeOfNavin (http://lifeofnav.in) and Gradient Ascent (http://navinpai.github.io/ga) and actively contributes back to the open source community as well. He is currently working at Bloomreach, a big data company focussed on the e-commerce sector. In his spare time, he enjoys writing about himself from the third person perspective.