[chatham]

Why are there so many physicists working with ANNs? Why the interest in ANNs in the field of high-energy physics? (examples: fnal.gov, CERN, etc)

Is there an implication that the principles of neural networks are analogous to those underlying physical reality? Sorry if that's froofy, but if that's not the implication then I really don't understand the interest.

[Orion]

Perhaps physics is simply benifitting from AI in the feild of data processing. I know not why neural nets in particular...
_________________
"The more numerous the laws, the more corrupt the state." --Tacitus

[DavidOlmsted]

I have noticed that also.

I think it is because of the mathematics in which conventional (addition based) neural networks are presented. It may also have something to do with the old and idea that neural networks bring order out of disorder just like the laws of physics brought order out of disorder. Originally the brain was thought to be just a mesh of neurons that somehow self-assembled with use. At the time genetic (biochemical) commands were ignored.
_________________
Click to go to my site at neurocomputing.org

[andy.goryachev]

Physics is about finding invariant representations. A discovery often happens when someone finds similarity - similar patterns - in different phenomena. I think any physicist immediately and intuitively understands that Jeff's HTM theory is a step in the right direction.

[FreeSynapse]

[dpkingma]

ANN's are popular because they can generate models of complex envorinments. Usually they still need lots of parameter tweaking (to prevent overfitting and other stuff), and the learning process still takes quite some time (depending on the environment complexity).

Note that HTM's can be classified as 'just another' form of ANN. What makes HTM's unique (as far as I know) is their unique configuration and the fact that the neocortex seems to use similar structures. That what makes them exciting imho.

[Nugenta]

I am physics-trained and activity involved in NN (HTM) research. Here is my answer; perhaps you might find it helpful?

From my perspective, the brain has evolved to solve two problems. First, it’s evolved to solve “the problem”. Second, it’s evolved to solve the problem with a minimum of energy. Right off the bat we are faced with questions of how the brain deals with computational efficiency, which gets us into the question of “what is a computation”. That topic inherently interests me, as a physicist. When you look into it deeper, you can see how a brain is a dissipative system that stores memory by constantly repairing that information. For example, if you go without oxygen for 10 minutes you will loose memories, but if you are in an ice bath during that time, you will likely be fine. This leads us into the realm of “probability of faults” or, stated another way “probability of state transitions”. This starts to scream of quantum mechanics to more you look into it. (the math is similar) When I look at the brain, its obvious that its working by iterating base rules, available at the neuronal level, over and over again. This is no different than postulated a fundament force or a new “particle interaction” and asking yourself how that rule, applied over and over again, results in the planets going around the earth, or the electrons probability of occupying states, etc. And lastly, as I think has been echoed by the others that have responded here, the brain does what a physicist is trained to do…it attempts to find the underlying causes in the world.

So, my opinion is that physicist like to study brains (or NN or HTMS, etc) because they have the tools to understand them. They have pre-learned a structure that seems to give them and edge in conceptualizing them.

~Alex

[FreeSynapse]

[Nugenta]

To be clear...I’m not saying that the brain works on quantum mechanical principles ( I really cant stand that theory ) but that the math and concepts having to do with states and state transitions are similar.

~Alex