Similarity search technology in images, music, and other multimedia content has been researched to death. This idea is not about research in any of those areas (I save that for work at Temple). The idea is simply an implementation of these techniques. Something like Google images that allows you to upload images and query based on similarity to the given image. Small-scale systems exist, but I have yet to find any that are as large as mainstream keyword-based image searches, such as Google Image Search. I’ve suggested this to Google when I was in their NYC office (I even gave them my BACH paper to suggest how they could do it for music!), but as far as I know, they still lack this feature (though they are joined by all of the other large search engines).

Large-scale query-by-humming systems already exist, so the lack of those isn’t a problem, but video could also benefit from such an approach (find video with this sound, find video with this frame, etc.). Images could be broken down using MPEG7 descriptors, time series analysis after linearization by a Hilbert curve, or vector quantization, among other techniques. Music could be broken down by a Fourier transform/power spectrum analysis; even the mood of the piece can be accurately predicted by this technique (according to the literature). Video search can be treated as a simple array of images and music (frames) and solved by the bagging the previous two methods.

Here’s an interesting idea, and one I’m in a position to test to boot!

Over several iterations of performing a simple motor task, an interesting pattern of activity occurs in the frontal cortex of the brain: the amount of activity diminishes with each iteration until a certain threshold is reached, indicating what appears to be motor learning behavior. It’s more or less linear, but I believe that the slope differs between subjects.

Now, my hypothesis is this: that the motor rate from a simple motor task could in fact be used to estimate the rapidity of motor learning in general. In other words, if I could stick you in an fMRI scanner for a few seconds and have you twiddle your thumbs, I could predict how fast you would be able to type or how long it would take you to learn to play the piano, for example.

But that’s only the beginning: that this is taking place in the frontal lobe rather than the cerebellum suggests that the processing may be somewhat unified with the process of normal cognitive learning, and thus may be a form of intelligence.

So can we build an IQ test from this? Maybe. I’m going to perform some quick experiments on data that we already have at the lab as soon as I can. Since I don’t have test data from the subjects, the initial analysis will be clustering, but if that succeeds, I may attempt to test them and perform regression.

If both of these hypotheses are true, Gardener’s theory of multiple intelligences may need to be amended.

The idea is simple, but should be effective: A class that meets solely to do research. It puts 30 or so minds on a project at once, and has everyone collaborating. Some people are naturally more effective in this environment too. It’s win/win.

If killing cancer cells off proves too difficult, undoing the process of malignant transformation and turning them back into benign tumor cells would still greatly reduce mortality from the disease.

I find it sort of pathetic that such little attention is being given to alternate immortality pathways in tumor cells. Come on, people! If you can subject the cells to aging, they will not divide indefinitely. If they do not divide indefinitely, they won’t survive very long. If tumors don’t survive very long, cancer no longer becomes a problem. Blocking telomerase is good (though there is some present in the body itself, especially in children, it is probably worth the tradeoff, and it would just be a temporary treatment measure anyway), but you need to block ALL of the ways cells have of circumventing the Hayflick limit.

Plus more telomerase research might dramatically extend the human lifespan, as we begin studying what is likely one major cause of aging. Not that I think we need even more people on the planet. If anything, we have too many already.

So let’s get some more people on it! It has to start with the funding, of course. Everyone follows the funding.

I’d help if I could, but I’m sort of… stuck.

It’s almost all self-study, but I have both a broad and deep knowledge of medicine. It’s an interesting field and one that I tend to absorb like a sponge. I can diagnose most diseases based on symptoms as well as any doctor. I know about protein regulation (it’s actually just a very complicated graph theory problem) and interaction. I can do gene sequencing, though I don’t think that would be a good area to put my skills to use. I can read most biology papers with ease. I easily know more about the diseases we study than any of the other computer scientists on the team, and I’ve occasionally surprised the biologists on our team as well. As a computer scientist in bioinformatics, I have an extensive ability to support my experiments with my own computational models. I have enough experience to read some types of medical images (though not nearly as well as a trained radiologist). And, as readers would see, I come up with all sorts of wacky treatment ideas both routinely and subconsciously, which means I’ll never run out of approaches.

SO SOMEBODY LET ME USE MY TALENT!

What I can’t do is use the equipment or get access to a lab. This is preventing me from doing experiments. I’m trying to find a bio course that will give me such access and enough training that I don’t blow stuff up or contaminate the lab’s cell lines, but this is Temple we’re talking about, not Polymath – I need to jump through more hoops than Shamu to enroll in a course outside of my major (OTOH, Temple is very well known in medicine; even more of a pity).

Maybe I care too much. I just hate being barred from the implementation of my ideas. Especially when I think those ideas are for the good of the very society that shuns them. I’m sometimes tempted to simply leave society on its own and go live the rest of my life on a farm somewhere, but I just can’t do it – I am going to create because I am the type that simply must create. It’s the “compulsion of ideas” that I speak of in my Treatise, but it’s of course a facet of one’s personality rather than the ideas themselves. Of course, I also argue that one’s perception of an idea forms the relative basis (as opposed to the absolute one) of the idea’s reality, so the compulsion is intrinsic to the combination of the idea and a receptive person in a sense.

Working on another classifier right now. This one needs to compare ROIs with individual codebooks. I’m not even sure we can meaningfully compare them, since the codeword indices don’t mean the same thing in different ROIs. I might need to use wavelets, which I had hoped to avoid. Tensor decomposition is the next step (and one I should get familiar with, because I’ll be working with it a lot in my dissertation).

I just keep getting older and the work keeps getting less exciting.

I’m going to continue my math research from where I left off soon as well. Maybe people would shut up about wondering how applicable a pure math result in number theory is if I make an attempt at proving GRH with it (Robin’s theorem lets me do that).

Probably not. Probably only if I solved it, which I probably can’t do with my current results, since my recurrence still fundamentally unrolls in accordance with the distribution of the primes that make up n. That never was the goal, though; I just like doing number theory. A lot.