JSNC

Went to JSNC today. As always, interesting stuff. Waydo's results were interesting. He achieved ~92% ROC curves for naive categorization of planes vs. cars vs. motorcycles vs. faces. Faces were categorized the best, as should be expected from such a dataset. He achieved even better ROC curves for 10 transformations of 10 faces each. His (Olshausen's) generative model was also interesting. If I get some downtime, it'd be worth looking at.

Izhikevich presented some fun results. Running a 1 second simulation of 10^11 neurons and 10^15 synapses. It took a 27 computer distributed network 50 days. Impressive, and the higher level image of the 1 second activation patterns is tres cool. You can see activation waves spreading over the neural gray matter. I'd love to see this narrowed down specifically using what we know about vision and visual anatomy. Hopefully, this is coming in the future. He projected a US$1mm computer system in the year 2016 should be able to do this computation in real time (using Moore's Law). Is Moore's Law still true?

A couple posters intrigued me, as well: Gepshtein's treatment of Gabor's uncertainty principle bears further looking into.