anhinga_anhinga: (Anhinga)
A year ago I posted about dataflow programming and linear models of computation:

http://anhinga-anhinga.livejournal.com/82757.html

It turns out that those dataflow matrix machines are a fairly powerful generalization of recurrent neural networks.Read more... )
anhinga_anhinga: (Default)
"The AGI conferences are the only major conference series devoted wholly and specifically to the creation of AI systems possessing general intelligence at the human level and ultimately beyond." This is a small conference (I expect around 200 people, give or take), and this year it comes in two parts, AGI-12 and AGI-Impacts:

http://agi-conference.org/2012/schedule/

http://www.winterintelligence.org/#calendar

The videos from the AGI-11 are available here:

http://agi-conf.org/2011/
anhinga_anhinga: (Default)
I recently found a very cool short paper (4 pages) which Dennis Gabor written 40 years ago.

It explains the math of associative holographic memory, which turns out to be very simple.

Because it is that simple, neither holograms, nor even waves and oscillations, are actually necessary to implement this scheme of associative memory.

D. Gabor, Associative Holographic Memories, IBM Journal of Research and Development, 13(2), 156-159 (1969). Abstract, PDF
anhinga_anhinga: (Default)
Numenta is founded by Jeff Hawkins, the author of "On Intelligence" (memory-prediction framework).

A research release of their "Numenta Platform for Intelligent Computing" is now available for Linux and Mac OS (free, targeted at sophisticated developers for the purpose of education and experimentation, no deployment rights).
anhinga_anhinga: (Default)
What might eventually turn to be the main event of 10th ICCNS a few weeks ago: Max Versace (a graduate student of Grossberg) finally did something to unify Grossberg's ART ("adaptive resonance theory") and spikes and synchronous oscillations.Read more... )
anhinga_anhinga: (Default)
More 10th ICCNS findings: Razvan Florian (Romania) presented a poster, "Reinforcement learning for spiking neural networks with modulated spike-timing-dependent plasticity". Basically, this modifies the reinforcement learning approach by Jonathan Baxter and Peter Bartlett for the situation with spikes. (One nice thing is that training for XOR works successfully in this case. I had difficulties with making the classical Baxter-Bartlett setup to work for XOR, either because of some principal problems, or because my hands grow from the wrong places.) His papers are here:

http://www.coneural.org/florian/

Meanwhile, Bostonians can see jaguar cubs at Stone Zoo in Stoneham. Here are some photos:

http://www.boston.com/news/local/massachusetts/gallery/052506_stonehamzoo_jaguars/
anhinga_anhinga: (Default)
Found an event-driven simulation package for modeling large networks of spiking neurons (GNU public license), have not tried it yet:
http://www.sccn.ucsd.edu/~arno/spikenet/papers.html
http://www.sccn.ucsd.edu/~arno/spikenet

Browsed the proceedings of "Advances in natural computation : first international conference, ICNC 2005, Changsha, China, August 27-29, 2005 : proceedings / Lipo Wang, Ke Chen, Yew Soon Ong (eds.)." 3 very think volume of Lecture Notes in Computer Science, 3610-3612 at a nearby library. Almost 4000 pages of papers mostly by Chinese authors. At least they still chose to write in English... (DBLP link: http://www.informatik.uni-trier.de/~ley/db/conf/icnc)

Profile

anhinga_anhinga: (Default)
anhinga_anhinga

July 2021

S M T W T F S
    123
45678910
11121314151617
18 192021222324
25262728293031

Syndicate

RSS Atom

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags
Page generated Jun. 20th, 2025 05:06 am
Powered by Dreamwidth Studios