Category Archives: private

News: First Deep Learning Course at MIT is a Hit with Students and Researchers

The following are notes for a potential news release on a Deep Learning for Self-Driving Cars course taught at MIT in January 2017. If any details are needed about any of the topics below, please contact Lex Fridman at:

News Item Summary

Deep learning is a collection of powerful artificial intelligence techniques that promises to revolutionize the way intelligent systems are able to leverage data to solve difficult real-world problems in driving, industrial robotics, healthcare, cybersecurity, speech recognition, automated translation, image captioning, high-frequency trading, weather forecasting, and many other application domains.

Lex Fridman taught the first course at MIT on deep learning: 6.S094 Deep Learning for Self-Driving Cars. It introduced the practice of deep learning through the applied theme of building a self-driving car (course website: http://selfdrivingcars.mit.edu). 250 students completed the course in-person and 7,000 completed it online, with over 12,000 students and researchers submitting neural network based agents to the traffic motion planning competition organized during the course.

Notes

  • IAP class taught in January 2017 by Lex Fridman
  • 250 MIT undegraduate and graduate students attended and completed all assignments for the course.
  • 7,000+ students globally registered and completed all assignments for the course.
  • The “DeepTraffic” (deep reinforcement learning) competition received 12,000+ submissions. 100’s of submissions are coming in every day.

Lecture Videos

The main lecture video is the Introduction to Deep Learning and Self-Driving Cars.

  • Lecture 1: Introduction to Deep Learning and Self-Driving Cars
    [ Slides ] – [ Lecture Video ]
  • Lecture 2: Deep Reinforcement Learning for Motion Planning
    [ Slides ] – [ Lecture Video ]
  • Lecture 3: Convolutional Neural Networks for End-to-End Learning of the Driving Task
    [ Slides ] – [ Lecture Video ]
  • Lecture 4: Recurrent Neural Networks for Steering through Time
    [ Slides ] – [ Lecture Video ]
  • Lecture 5: Deep Learning for Human-Centered Semi-Autonomous Vehicles
    [ Slides ] – [ Lecture Video ]

Interesting Images and Videos

This Dropbox directory contains interesting images, gif animations, and videos that may be useful as part of a news article on this topic. Here is a description of each:

  • deepcars_stats.png – shows the statistics on the 5 lecture videos, which includes 170k+ views and 2+ million minutes watched.
  • lecture_thumbnail.jpg – snapshot of Lex teaching one of the lectures.
  • deep_traffic_animation.gif – gif clip of the DeepTraffic competition. This competition involved building a deep reinforcement learning agent that weaved around cars through traffic with the goal of achieving that highest average speed without colliding with any of the other cars.
  • deep_traffic_video.mp4 – mp4 clip version of the DeepTraffic competition.
  • deep_tesla_video.mp4 – mp4 clip version of the DeepTesla simulation. This simulation involved building a convolutional neural network for steering a Tesla vehicle based on a single camera video stream of the forward roadway.

Some of these images and videos are included below.

The following is deepcars_stats.png which shows the statistics on the 5 lecture videos, which includes 170k+ views and 2+ million minutes watched:

deepcars-stats

The following is deep_traffic_video.mp4 which is a clip of the DeepTraffic competition. This competition involved building a deep reinforcement learning agent that weaved around cars through traffic with the goal of achieving that highest average speed without colliding with any of the other cars.

 

The following is lecture_thumbnail.jpg which is a snapshot of Lex teaching one of the lectures.

lecture_thumbnail

The following is deep_tesla_video.mp4 which is an mp4 clip version of the DeepTesla simulation. This simulation involved building a convolutional neural network for steering a Tesla vehicle based on a single camera video stream of the forward roadway.