Wednesday, February 15, 2017

AI Startup Gamalon

Read: Bayesian Program Synthesis (BPS). The technology is such that it writes and rewrites its own Bayesian programs. 

Sounds like Data Equilibrium and AODiagrams. Check it out on Source Forge
Added:  I was wrong, it is much more like Narwhal.
Added: Wrong again. I watched a great TED talk from the founder of Gamalon at 2:30 AM and then went back to bed to think about it. That fellow understands perfectly well that models, which he calls "stories", are critical in forming a machine learning solution. He also vindicates my impression that (1) AI is not new but is being amp'ed up by powerful machinery and data bases; and (2) that 90% of the AI startups [of which there are many in the Boston area these days] do not understand the importance of "stories" and, by implication, those startups will fail.
Good stuff! He is right. He also tried to introduce the principle of using the goodness of fit as a weighting factor in a Bayesian Network. Not sure what he means there.
Added: In Bayesian networks it is the connections between nodes that get weighted, so to make sense of what he said, is to put narratives in the role of connections between nodes of a graph. But you know, there are more than one way these nodes can be connected in different narrative structures. Still not sure I get it.

No comments:

Post a Comment