clevercode

Agreed. Also, sentences like "I mastered the ability to combine several strategies, each with varying strengths and weaknesses, highlighting the strengths and nullifying the weaknesses" are no good. What strengths? What weaknesses? When I read something like that, it says to me (or perhaps screams?) that nothing substantial was actually accomplished. Definitely in need of rewording to make things far more concrete.

One recent development in the field of artificial neural networks is "Deep Learning": http://www.youtube.com/watch?v=VdIURAu1-aU&feature=playe...!

True, but that fact only becomes useful after you already have it running on a virtual machine. Chicken-and-egg problem.

Recreating the source code by itself does not, in this case, constitute a "model". In order to actually run that binary, you also need to create a virtual computer for it to inhabit.

By "looping", I'm assuming you mean "recurrent"? If so, you might be interested these:

http://www.idsia.ch/~juergen/rnn.html

http://www.lsm.tugraz.at/learning/framework.html

http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.31.6...

The first study you link to does not appear to address the correlation/causation issue whatsoever. Specifically: "The study acknowledges that people who regularly run marathons, like many of the study's participants, may be genetically predisposed to running long distances."

The second link describing neurogenesis stimulation is more interesting due to the fact that they are tracking actual changes in the participants over the course of the study.

For the record, I do not doubt for a moment that exercise has numerous positive benefits, both physically and mentally.

I wonder how much of this is simply due to poker's recent surge in popularity, thus creating a much larger pool of potential talent to draw from.

A possible (somewhat sci-fi) future extension to this idea would be directly monitoring the brain of the subject, in conjunction with eye movements, in an attempt to quickly detect a subconscious signal of 'recognition' while scanning the computer-generated faces. Might be that certain signals in the brainwave (e.g. the P300 or something similar: http://en.wikipedia.org/wiki/P300_%28neuroscience%29 ) could be used for this purpose.

I heartily agree with your skepticism.

In my opinion, rather than attempting to simulate progressively larger brains, we should instead focus on really nailing the details of smaller ones, perhaps starting with insects.

As an example, Portia jumping spiders display very complex hunting behaviors, including trial-and-error learning in novel circumstances. Even with tiny eyes, they have the visual acuity of a cat. However, unlike a cat, they are only able to see a small portion of the visual field at a time, and must therefore scan the environment, keeping the rest of what has been scanned in memory. Portia spiders are known to spend up to an hour "analyzing" their environment before deciding on a course of action! Their sharp eyesight allows them to recognize the species of the spider they are currently stalking. One common tactic is to pluck deceptively at the web of their victim in order to manipulate it into reacting and moving into a more favorable position before the attack. Different species of spiders will respond in different ways, and so the Portia learn to intelligently apply different patterns of plucking to fit the circumstances. When encountering an unknown type of spider, Portia will try various experimental plucking motions, the successful results of which will be remembered and reused correctly in future attacks.

This amazingly rich and adaptive behavior it somehow manages to produce with a mere 500k neurons; about 2000 times less than the number used by Blue Gene in the experiment.

Interesting idea! I wonder if that's why I always feel so much more awake and focused after taking a shower?

I agree that analysis of neural networks is not easy, but there are certainly ways to do it:

there is no way to find out the relative significance of the input variables and how they affect the output response

Calculate partial derivatives of inputs w.r.t. outputs.

there isn't a way to cluster similar cases together to analyze what rules bind different decisions together

PCA analysis of activations of the hidden layer. Could also do k-means clustering of activations of hidden layer.

The random forest classifier looks interesting, I'll have to investigate that. Any suggestions for papers/tutorials?

What worked for me was proper diet and exercise. I used to suffer from pretty bad panic attacks that would occur randomly, and it was really debilitating. The thing that really helped me was taking up jogging. Some form of daily exercise is essential. Also, cutting out caffeine completely helped a lot.

Good luck - it might take awhile to get back to 'normal', but keep at it!

I agree with you in principle, however, their simulations are approximations, not identical.

From their website FAQ:

Q: Do you believe a computer can ever be an exact simulation of the human brain?

A: This is neither likely nor necessary. It will be very difficult because, in the brain, every molecule is a powerful computer and we would need to simulate the structure and function of trillions upon trillions of molecules as well as all the rules that govern how they interact. You would literally need computers that are trillions of times bigger and faster than anything existing today.

Because this will only be an approximation, it means that certain hard-to-define grammatical properties of spike-trains (e.g. "neural codes" as opposed to "rate codes") may be completely lacking in the simulation. Do these "neural codes" exist? At this point we do not know.

If it turns out that the brain does utilize complex neural codes, and that they are not being accounted for in the simulation, then the simulation can be no more than gibberish which happens to share many low-order statistical features with the real thing.

To illustrate this point, here are some examples of pseudo-English sentences automatically generated by a purely statistical model (n-gram model where n=2):

Richard Beesemyers, formerly raised sagging candidate to the Friday officially forecast at the project.

A hearing appeals.

That bank handles most notable exceptions to buy time for reconsideration Wednesday in decades, is the B& Coopers, to discuss international theme for the violin away as the wages of the power plant near classic chemise.

It could have been observing the 65 to back Tuesday.

And fourth, there is a full report, Wagner said.

I'm very curious if, during this demonstration, you were actually allowed to directly interact with the system or not?

I got 97% on 'Particle Pro'

You are correct, I should have read more carefully.

Nevermind, you cited it in your paper... :P

Neat stuff! One thing you might want to look into (if you haven't already) is Probabilistic Incremental Program Evolution (PIPE) by Schmidhuber.