Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

TensorKart: Self-Driving MarioKart With TensorFlow

358 views

Published on

Presented at FITC Toronto 2017
More info at http://fitc.ca/event/to17/

Presented by Kevin Hughes, Shopify
Overview

TensorKart is an end-to-end deep learning demonstration about training an AI to play MarioKart 64 with TensorFlow. Using the same technique as Google’s self-driving cars this talk will go through the process step by step and explain it in an approachable way. By the end of the session you’ll see an AI playing MarioKart 64 and understand how it works regardless of previous experience in deep learning or programming.

Objective

to bring deep learning to a wider audience in an approachable way

Target Audience

Anyone interested in artificial intelligence or self driving cars

Assumed Audience Knowledge

Some basic programming knowledge or at least interest is all that is required to enjoy this talk

Three Things Audience Members Will Learn

What TensorFlow and deep learning are all about
How to approach an AI problem using deep learning
A deeper appreciation for how real self driving cars might work

Published in: Internet
  • Be the first to comment

  • Be the first to like this

TensorKart: Self-Driving MarioKart With TensorFlow

  1. 1. TENSORKART SELF-DRIVING MARIO KART WITH TENSORFLOW
  2. 2. HELLO! ▸ Senior Developer at Shopify ▸ Previously an AI researcher at Queen’s University
  3. 3. WHAT IS DEEP LEARNING?
  4. 4. MACHINE LEARNING..?
  5. 5. source: https://upxacademy.com/spam-alert-machine-learning-is-filtering-your-emails/
  6. 6. LAZY
  7. 7. source: https://www.kaggle.com/benhamner/nips-2015-papers/discussion/22778
  8. 8. source: http://www.butleranalytics.com/neural-network-data-mining-explained/
  9. 9. Source: http://synapsebristol.blogspot.ca/2013/02/the-ai-lab-brain-computer-interfaces.html
  10. 10. TENSORKART
  11. 11. THE PROJECT ▸ Use Deep Learning to train an AI to play Mario Kart 64 ▸ Document the entire process
  12. 12. STEP 1 - TRAINING DATA ▸ Save images of an N64 emulator window ▸ Record the joystick input from a user at the same time
  13. 13. STEP 2 - BUILDING THE MODEL ▸ I used TensorFlow - the latest Deep Learning Framework from Google Nvidia Autopilot Model
  14. 14. STEP 3 - TRAINING ▸ Collect all the training data and feed it to the model ▸ Its an optimization problem - TensorFlow solves for the smallest error across the dataset TensorFlow training
  15. 15. STEP 4 - PLAYING ▸ How to connect the AI back to Mario Kart? ▸ Fake joystick
  16. 16. DOWN THE RABBIT HOLE
  17. 17. n
  18. 18. GO TIME
  19. 19. BACK TO THE DATA ▸ Remove corrupted screenshots ▸ Drive smoother ▸ Play more Mario Kart
  20. 20. TAKE 2
  21. 21. WHY?
  22. 22. THANKS! ▸ Read the full post at kevinhughes.ca/blog/tensor-kart ▸ Checkout the source on Github kevinhughes27/TensorKart ▸ Follow me on Twitter @kevinhughes27

×