2013 Fall Hackathon Outcome
Last weekend’s hackathon was a lot of fun. Thanks to everyone for coming out. We gave several educational presentations and held some working sessions. The participation from the attendees was outstanding.
A panoramic view of the hackathon space at Tagged.
Thanks again to Tagged, who generously donated some of their workspace to us for the event. The Tagged offices are very cool, and in a great part of San Francisco.
Jeff Hawkins talking about sensor-motor integration in the neocortex.
We had over 50 people attend the hackathon from across the United States and Europe, most of whom stuck around for the demos after the 30-hour hack period. In all, we had 7 impressive demos (see below)! Only a couple were natural language processing hacks, but that’s okay. It was great to see NuPIC users making their ideas a reality in whichever direction they chose to apply it.
Hackathon attendees watching Subutai Ahmad’s CLA Deep Dive presentation.
“It was easily the best hackathon I have attended this year.” —Jeff Fohl
“There was fantastic energy in the room, something that only reinforced my sense that Numenta is on to something really big.” —Johann Schleier-Smith
“Very informative and spiked my interest in machine learning and NuPIC.” —Venkat Venkataraju
“Great event, convinced to jump in and start experimenting. Numenta seems to be going in the right direction.” —Brian Eppert
“Great experience. Intelligent people, insightful talks. I was totally new to NuPIC and although I didn’t complete a project, I feel like I’ve learned a lot about the technology and the whole approach.” —Nikolay Samusik
“I think we had a nice mix of people at the event, e.g. from those focused on neuroscience biological aspects, to those just getting their feet wet, a few hackers applying NuPIC to some applications, and the great staff. Great new introductions, good conversations and informational sessions. Definitely an event worth attending!” —Ari Kamlani
Hacking in progress!
We tried to record everything we could, which included all the sessions (minus one because of a technical glitch). My apologies for the poor audio and video in some of these. We had a wireless microphone failure because of unanticipated RF interference. We attempted to live-stream everything on YouTube via Google Hangout, which meant we sacrificed quality. For future hackathons, I plan to have better tools in place.
Jeff and I got the event rolling. This includes protocol information, thank-yous, and guidance for hackers.
Matthew Taylor had done some NLP work before the hackathon in order to help others get started doing some work with NuPIC, NLP, and especially the CEPT API for word SDRs (http://cortical.io). In this presentation, I present some of the initial progress I made.
Subutai Ahmad, Grok/Numenta VP of Engineering, detailed some aspects of the CLA. He discussed an interesting property of SDR’s affecting temporal pooling and hierarchies. The interactive session included a lot of Q&A
Most of the changes on our senses are due, wholly or in part, to our own actions. Therefore, the neocortex learns a sensory-motor model of the world. In this talk, Jeff discussed some of what we know about how the neocortex generates behavior and how we might construct a simple sensory-motor system based on the Cortical Learning Algorithm. It is a good starting point for anyone wanting to add motor output to the CLA.
This was an informal question and answer session between some hackathon attendees and Jeff and Subutai.
Francisco Webber talked about his service that provides text comparison and association services, as well as SDRs for English words. He described in great detail how these SDRs are created, and talked about how they are being used by Phase 6 in their effort to apply this technology to better English education opportunities.
Matt compiled and ran NuPIC on a Raspberry Pi, and wired up some wheels, a controller, and light-sensitive sensors to help drive it. While he programmed it to continuously steer towards the brightest light source, he used NuPIC’s anomaly scores of the light input from right and left to stop the vehicle when it wandered into unexpected territory.
- Video: http://www.youtube.com/watch?v=X4XjYXFRIAQ&start=390
- By: Matt Keith
NuPIC on ARM7
Ari compiled NuPIC into ARM 7, ran it on a Chromebook, and talked about using NuPIC to predict the signal strength of public wifi access points.
- Video: http://www.youtube.com/watch?v=X4XjYXFRIAQ&start=990
- By: Ari Kamlani
This hack intends to give the CLA an ability to evaluate hypothetical input. Imagine that you want to assess (predict) what the consequences of some action (input) would be. You may have a set of possible inputs (as alternatives), and want to evaluate which would lead to the best (predicted) outcome. The demonstration will include mice, a maze, and some cheese.
- Video: http://www.youtube.com/watch?v=X4XjYXFRIAQ&start=1970
- Code: https://github.com/EronWright/nupic-mouse
- By: Eron Wright
Continuing with the NLP project I introduced in the Kickoff presentation, I created bitmap images of the word SDRs predicted by the NuPIC temporal pooler and compared them to the words the CEPT API associated them with.
- Video: http://www.youtube.com/watch?v=X4XjYXFRIAQ&start=3240
- Code: https://github.com/rhyolight/nupic_nlp
- By: Matt Taylor
This team taught the CLA to learn to control a quadcopter. Their objective was to pilot the quadcopter by controlling its vertical speed to fly to a desired altitude and hover there. The supervised learning approach involved having the CLA watch an expert flyer and learn those sequences, and use its prediction of those sequences to directly control the copter in the face of noise.
- Video: http://www.youtube.com/watch?v=X4XjYXFRIAQ&start=4080
- Code: https://github.com/chetan51/pilot
- By: Chetan Surpur, Jay Gokhale, Josh Rose, Xuyen On, Eron Wright
Spatial Pooler Visualization
Ian showed a very enlightening visualization of a live spatial pooler being fed parts of an image, one piece at a time. His hack displayed the cellular activation given different SP settings, as well as visualization of the column permanences.
- Video: http://www.youtube.com/watch?v=X4XjYXFRIAQ&start=6110
- Code: https://github.com/iandanforth/spviewer
- By: Ian Danforth
What Does the Fox Eat?
Subutai used the word SDR association framework defined in nupic_nlp to associate three word phrases, teaching it sentences like “elephants eat leaves”, “dogs like sleep”, and “cows eat grass”. After some training, Subutai asks NuPIC, "What does the fox eat?" Find out the answer by watching the video!
- Video: http://www.youtube.com/watch?v=X4XjYXFRIAQ&start=7084
- Code: https://github.com/numenta/nupic.nlp-examples
- By: Subutai Ahmad
Lastly, our CEO and Chairman of the Grok/Numenta Board, Donna Dubinsky, said a few inspirational words at the end of the hackathon. Thanks, Donna!
This was a really fun event to plan, and it emphasizes what a great community we have with NuPIC. Thank you everyone, for helping make this open source project what it is today. I am looking forward to planning our next hackathon this Spring!