I like the ANN analogy more than the JIT one... though perhaps my understanding of either is flawed. The JIT analogy is stronger, I think, because ANNs aren't (?) typically capable of multi-domain classification (right?). The extent to which they can operate over a space on which they're not trained is very limited. But the JIT, because its ultimate input (turing complete languages) and output (general purpose computers) are universally expressive, can apply across a huge number of domains. It seems like the sizes of the fan-in and fan-out for compilers are huge compared to those for ANNs.
On 11/08/2016 07:45 AM, Marcus Daniels wrote: > A neural net trained to discriminate between nuances in one environment (H) > would need to get re-trained (or I'd say untrained) to the D environment. > The signals in H type environments are higher dimensional, coupled, and > non-linear compared to the D environment which is made up of many more > independent and simpler hazards. With finite resources, I expect the > H-specialized M agent apparatus needs to be torn-down to make room for > constant bombardment of D-world wild dogs. Not really interpreted vs. > compiled, more like a Java hotspot JIT that is constantly refining to the > environment. -- ␦glen? ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com FRIAM-COMIC http://friam-comic.blogspot.com/ by Dr. Strangelove
