neurodudes

The symbol grounding problem (apparently) is: assuming that someone’s “cognitive” levels of mind work in terms of symbols, how to design the interface of the symbolic levels with the low-level sensory-motor systems?

Jun Tani suggests a “recurrent neural network with parametric biases” (RNNPB).

I haven’t had time to read further yet, but it looks very interesting so I’m passing it along now lest it get lost. When I get around to it I’ll post an update that summarizes what RNNPBs are and precisely how they interface symbol computation with lower-level systems. I may attend the talk (which is tomorrow).

Read on for an abstract from Jun Tani’s talk.
(more…)

btw, we’ve gotten lots of spam comments. our solution is to use an automated spam blocker to screen comments. we’ve had to block the brand names of many drugs, as well as lots of words related to gambling. If you try to submit a comment and it is rejected due to “questionable content”, then try rephrasing the comment to not use an “spamy” words, and using the generic name for any drugs you mention. Or, email the administrators, and we’ll help.
(more…)

University of Florida has an Oct 21 press release about Thomas DeMarse’s work in hybrots, that is, cultured neurons interacting with a computer via an MEA to control an avatar in a simulated world. In this case, the neurons are flying a plane in a flight simulation program.

It looks from the press release that he has got the animat to actually learn to fly a plane (i.e. got input about the simulated environment and controlled a simulator joystick to keep the plane steady!

full article

Unfortunately he doesn’t describe what kind of learning rule/feedback was used, so I guess we’ll have to wait for the paper to see how much excitement is justified.

P.S. there’s also a lively discussion on SlashDot about this press release.

I’d be interested in hearing which parts of SFN people are planning to go to. I’m especially looking for computational and theoretically oriented stuff.

You can see stuff that I am thinking about going to here.

Btw, both Neville and I will be in San Diego for SFN (well, I’m always in San Diego, but Neville’s visiting for the conference).

From Neuron (Sep 2), an interesting combination of mathematical analysis and proof with biology:

Synaptic Connectivity and Neuronal Morphology
Two Sides of the Same Coin
Dmitri B. Chklovskii

Neurons often possess elaborate axonal and dendritic arbors. Why do these arbors exist and what determines their form and dimensions? To answer these questions, I consider the wiring up of a large highly interconnected neuronal network, such as the cortical column. Implementation of such a network in the allotted volume requires all the salient features of neuronal morphology: the existence of branching dendrites and axons and the presence of dendritic spines. Therefore, the requirement of high interconnectivity is, in itself, sufficient to account for the existence of these features. Moreover, the actual lengths of axons and dendrites are close to the smallest possible length for a given interconnectivity, arguing that high interconnectivity is essential for cortical function.

Article Outline

? Introduction
? Results

? Design I: Point-to-Point Axons
? Design II: Branching Axons
? Design III: Branching Axons and Dendrites
? Design IV: Branching Axons and Spiny Dendrites
? Proof of Design Optimality
? Comparison with Experiment

? Discussion
? Acknowledgements
? Appendix
? Note Added in Proof
? References

Full article

“These lectures were recorded at the Obidos Advanced Course in Computational Neuroscience 2003. This is a yearly summerschool in neuroscience and related areas. We are grateful for the speakers’ permisssion to use their material.”

Haven’t checked these out yet but several look interesting. Larry Abbott has a lecture on spike-timing dependent plasticity, Alain Destexhe has a lecture on synaptic noise in the cortex, and Dan Wolpert has one on sensorimotor learning.

Here’s the link to the lectures.

This looks cool; a machine learning algorithm for learning to cluster, based on training data. You specify a set of attributes which can be computed on any given local neighborhood (nearby subset of points), and then the algorithm learns how to use these “features” to cluster.

For more info, search http://www.psi.toronto.edu/ml.html for “learning to cluster”.

Training data:
Test data:
Clustering produced by the algorithm:
The correct answer:

Recent work in Nature from a group in Germany shows the importance of temporal order in determing whether a stimulus is considered aversive or appetitive. Briefly, fruit flies were given paired odors and electric shocks with a varying interval between the two events; when the odor preceeded the shock, it became an aversive stimulus. When shock preceeded odor, the odor became appetitive.

It amazes me that with the recent interest in reinforcement learning, so many of the “cutting-edge experiments” look awfully similar to the behaviorist literature of almost 100 years ago! Although I don’t know the literature well myself, I am positive that someone must have taken Pavlov’s famous experiment and reversed the temporal order (maybe not for fruit flies but for dogs or pigeons perhaps). If anyone out there knows of something, please post it below in the comments.

Also, there is some striking similarity with MM Poo’s work with growth cone guidance in which a chemotactic factor is combing with electrical stimulation. In that work, certain patters of electrical stimulation were able to reverse the actions of particular chemotrophins from attraction to repulsion or vice versa.

Full article from the German group can be found after the jump.
(more…)

Amazing results. I hope this serves as a wake-up call to those who still haven’t realized the incredible advances that are being made possible through the use of embryonic stem cells. From the July 27, 2004 issue of PNAS:

Transplanted human fetal neural stem cells survive, migrate, and differentiate in ischemic rat cerebral cortex
S. Kelly, T. M. Bliss, A. K. Shah, G. H. Sun, M. Ma, W. C. Foo, J. Masel, M. A. Yenari, I. L. Weissman, N. Uchida, T. Palmer and G. K. Steinberg

Full article

We characterize the survival, migration, and differentiation of human neurospheres derived from CNS stem cells transplanted into the ischemic cortex of rats 7 days after distal middle cerebral artery occlusion. Transplanted neurospheres survived robustly in naive and ischemic brains 4 wk posttransplant. Survival was influenced by proximity of the graft to the stroke lesion and was negatively correlated with the number of IB4-positive inflammatory cells. Targeted migration of the human cells was seen in ischemic animals, with many human cells migrating long distances ({approx}1.2 mm) predominantly toward the lesion; in naive rats, cells migrated radially from the injection site in smaller number and over shorter distances (0.2 mm). The majority of migrating cells in ischemic rats had a neuronal phenotype. Migrating cells between the graft and the lesion expressed the neuroblast marker doublecortin, whereas human cells at the lesion border expressed the immature neuronal marker {beta}-tubulin, although a small percentage of cells at the lesion border also expressed glial fibrillary acid protein (GFAP). Thus, transplanted human CNS (hCNS)-derived neurospheres survived robustly in naive and ischemic brains, and the microenvironment influenced their migration and fate.

almost offtopic, but neat; here’s a course on “DNA and Molecular Computation”, with a list of papers such as “Chemical Kinetics is Turing Universal”. I didn’t know so much work had been done on this stuff already! cool!

http://www.cs.caltech.edu/courses/cns288/

(btw, if you’re interested in reading one of the actual articles, here’s Chemical implementation of finite-state machines, which goes into more (computational) detail than “Chemical Kinetics is Turing Universal” actually does).