MIT – Social Blog

As BIM-student, it is very likely that you are interested in topics like coding, Deep Learning, Artificial Intelligence, Machine Learning, human-robotic interaction, or Autonomous Vehicles. If by any chance you also enjoy listening to podcasts, you might be in luck:

I highly suggest you to check out the Lex Fridman Podcast.

LexFridman

Lex Fridman is an AI research scientist at the Massachusetts Institute of Technology, often better known as MIT. He works on developing deep learning approaches to human sensing, scene understanding, and human-AI interaction. He is particularly interested in applying these technologies in the field of Autonomous Driving.

LexFridmanTeaching

If you know the Joe Rogan Experience, you likely are already familiar with Lex. Having worked for both Google and Tesla, Lex Fridman understands the business application of digital technologies. He uses his podcast to share this knowledge with his audience and discusses his fascination with a variety of interesting guests. This can be particularly interesting for us as Business Information Management students, as we also form the future bridge between business ventures and technological innovation. The podcast discusses similar topics like we get taught in class, sometimes going more in depth, with international research experts in those particular fields.

If you enjoy podcasts, these are some examples of Lex Fridman Podcast episodes that I highly recommend you to give a listen as a BIM-student:
RecommendedEpisodes

Episode #31 with George Hotz: Comma.ai, OpenPilot, Autonomous Vehicles.
Famous security hacker. First to hack the iPhone. First to hack the PlayStation 3. Started Comma.ai to create his own vehicle automation machine learning application. Wants to offer a $1000 automotive driving application, which drivers can use on their phone.

Episode #49 with Elon Musk: Neuralink, AI, Autopilot, and the Pale Blue Dot.
Elon Musk. Tech entrepreneur and founder of companies like Tesla, SpaceX, PayPal, Neuralink, OpenAI, and The Boring Company.

Episode #114 with Russ Tedrake: Underactuated Robotics.
Professor of Electrical Engineering and Computer Science, Aeronautics and Astronautics, and Mechanical Engineering at MIT.

Episode #120 with François Chollet: Measures of Intelligence.
French Software Engineer and researcher in Artificial Intelligence, who works for Google. Author of Keras – keras.io – a leading deep learning framework for Python, used by organisations such as CERN, Microsoft Research, NASA, Netflix, Yelp, Uber, and Google.

These were just several examples of episodes that I enjoyed myself.

The benefit of a podcast is that you can listen it basically anywhere, and can stop listening at any time. If you are not familiar with podcasts yet or with the listening experience they offer, maybe the Lex Fridman Podcast could be your first step into this experience.

You can find the episodes of the Lex Fridman Podcast here: https://lexfridman.com/podcast/

Or check out Lex Fridman’s Youtube channel here: https://www.youtube.com/user/lexfridman

The above sources have been used as sources for this post.

5/5 (7)

No ratings yet.

Just in time for Halloween MIT has started a project to generate scary imagery using an image processing algorithm based on artificial intelligence, specifically through deep learning. The algorithm can recognize and generate scary faces and make places look scarily. It is interesting to think about the purpose of this experiment, because can machines make us scared? The result is scarily accurate.

The image generation works through several steps. First it uses deep learning to generate completely new faces. Subsequently a filter with extra imagery is applied to scarify to the faces. After that the system learns from a voting system about the extent of scariness. The system that is behind this algorithm is based on unsupervised learning with convolutional networks (CNNs) and convolutional generative adversarial networks (DCGANs). If you want to read more about this, find the very interesting and recent conference paper by Chintala, Metz and Radford (2016) at https://arxiv.org/abs/1511.06434.

The cool thing is that you can help training the algorithm by going to http://nightmare.mit.edu/faces. After trying it myself I do not think that the machine still needs a big increment in learning anymore as the pictures already look quite scary. What do you think?

It is also interesting to think about the possibilities of this development in the future. Chintala, Metz and Redford (2016) imagine future applicabilities in video prediction as well as for audio. I imagine implications in for example virtual reality in which the environment is automatically generated with the intent to make you scared. The system could learn from how you react on the input and personalize the algorithm to make you even more scared. Moreover just think about how this can be applied to theme parks to make those experiences even more immersive!

All in all this development is a step towards a scary but also very exciting future!

If you have any comments or other ideas for similar applications in artificial intelligence, let me know! If you want to read more about the algorithm go to http://nightmare.mit.edu.

http://boingboing.net/2016/10/23/using-machine-learning-to-auto.html

http://nightmare.mit.edu/faces

http://nightmare.mit.edu

https://arxiv.org/abs/1511.06434

Image from http://boingboing.net/2016/10/23/using-machine-learning-to-auto.html

7

Please rate this

23

Please rate this