Tag: machine learning

Posted on
by

It’s been a little over a year since my last post. Since you’re reading my personal blog you might be interested to know what I have been diving into in the meantime. Here’s a short list:

  • Artificial intelligence and machine learning. I really think that AI systems need to be free to learn in order to become really useful. I was excited by Jeff Clune’s work on open-ended learning. I heard about his work through the TWIML podcast. Jeff Clune worked on this amazing project called POET where the AI system created obstacle courses and used them to train its parkour skills. I also got into Neural Operators which are a class of neural networks that map between functions instead of between vectors like normal neural nets. The sound like they are great for physics simulations.
  • Personal knowledge graphs. I switched from Evernote to Obsidian and Joplin. I adapted my task management workflow using this post, by Ruben Berenguel, as inspiration. There was a lot of fiddling and learning involved, but I like how all my notes are in markdown text files now. I feel more free.
  • Joined Mastodon social network. I’ve been really enjoying it. If you haven’t heard of it, it is a decentralized social network kind of like Twitter. It is part of the Fediverse which is a collection of different decentralized social networks that all interoperate together. Here’s my Mastodon profile, https://raphus.social/@davidruffner. Come check out the Fediverse!
  • Free and Open Source Software (FOSS). I did a deep dive into it. I learned a lot from the Libre Lounge podcast, hosted by Serge Wroclawski, and the FOSS and Crafts podcast, hosted by Morgan Lemmer-Webber and Christine Lemmer-Webber. I’ve been slowly transitioning my apps to open source alternatives.
  • I’ve been dabbling in the Elixir and Scheme programming languages. I took the Hands-On Guile Scheme for Beginners class and enjoyed it. I’m still looking to get better at the Actor Model, but have definitely gotten some exposure to functional programming.
  • I did a dive into Category Theory. I think someone shared about it on Mastodon, and I was blown away by this series of lectures by Bartosz Milewski.
  • I’m did a deep dive into Marshall Rosenberg’s Non-Violent Communication. It has helped me to grapple with what is going on between Israel and Gaza, and closer to home as helped with interpersonal conflicts. I’m taking the Bigbie Method intro course and enjoying it so far.
  • FOSS video games. I’ve been having a lot of fun finding free and open source video games to play myself and to share with my two sons (5 and 2). Here are some of the top games so far:
    • SuperTuxKart. We all love it! I got some xBox360 controllers so we can play on the computer as if it is a gaming console. It’s fun finding fan-made tracks and cars too.
    • Sonic Robo Blast 2 a fun retro 3D Sonic game.
    • 0 A.D. An excellent real time strategy game like Age of Empires

Posted on
by

My colleague, Jarek Blusewicz, recently told me about autonomous RC cars and how they can be a fun project for playing around with AI and robotics. I didn’t know that autonomous RC cars were a thing, but it turns out that lots of individuals and teams are playing around with them, and doing awesome work. I found a video of a RC car autonomously drifting around in controlled circles. It is amazing to watch, especially when the camera zooms in on the front wheels steering back and forth to maintain the drift as the car slides over the surface. It looks like a professional driver, but the car is completely controlled by a computer.

Researchers at Georgia Tech created an autonomous rally RC car for “aggressive off-road driving”. They really mean aggressive! The car hits a turn at almost full speed and then power slides through it while kicking up clouds of dirt. Here’s a picture of their AutoRally car in action.

autorally_platform_header
Credit: AutoRally: An Open Platform for Aggressive Autonomous Driving. Brian Goldfain, Paul Drews, Changxi You, Matthew Barulic, Orlin Velev, Panagiotis Tsiotras, James M. Rehg. Control Systems Magazine (CSM), 2019.

The best part about the Georgia Tech work is that it is all open source, with a site on Github including detailed instructions about how to build the car (it wouldn’t be cheap, but could be great for a group or organization).

The AutoRally car is controlled with a reinforcement learning algorithm, which I have been finding to be widespread for controlling robots in general. It’s something that I would like to learn more about and play with.

 

Posted on
by

Since finishing grad school, I have been diving from physics into coding. It turns out that it probably is all the same thing at the end of the day! I came across a paper that discovered an exact mapping between the renormalization group, a central concept in modern physics, and deep learning, the latest and greatest machine learning algorithm. It’s crazy but makes sense. Both the physics idea and deep learning rely on looking at a system on a succession of different levels. This high level view of a system helps you to see the big picture. That is useful for self-driving cars and for finding phase transitions in condensed matter systems.

I am excited about the potential of machine learning and AI. It’s cool to think that my various seemingly different interests, could turn out to overlap in productive ways.

What if there are other hidden connections between totally different fields?