Category Archives: Research

Last

In the wake of what may have been the closest thing I’ve ever taken to an all-nighter (I had told my advisor that I would have the Wavelet paper by tonight and I would have it by tonight if I had to stay awake all night to complete it… I really try hard to keep my word), I took a look at the “last” logs on the server after finishing, curious about how everyone else’s conference projects are coming along:

(Names changed and IPs removed to protect the innocent… and the server)

[michael@aristotle Motor Code]$ last
michael pts/1 Tue Jan 8 21:49 still logged in
michael pts/1 Tue Jan 8 14:47 – 21:35 (06:47)
michael pts/1 Mon Jan 7 16:05 – 19:47 (03:41)
tom pts/4 Fri Jan 4 13:08 – 13:09 (00:00)
tom pts/4 Fri Jan 4 13:00 – 13:08 (00:07)
michael pts/1 Fri Jan 4 11:31 – 16:40 (05:08)
wang pts/1 Fri Jan 4 09:14 – 09:17 (00:03)
michael pts/1 Thu Jan 3 14:54 – 15:28 (00:33)
michael pts/4 Wed Jan 2 14:02 – 14:11 (00:09)
michael pts/1 Wed Jan 2 14:02 – 14:12 (00:10)

“Tom” helps maintain the server and “wang” is an alumnus who graduated last year (but still logs on very frequently). Rather than assume I’m the only one using the server regularly (because I’m not most of the year), I’m assuming that everyone else just has later conference deadlines.

I’ve been logged on around 18 hours this week, which doesn’t sound like a lot but means I’ve probably been doing research for closer to 36. By most standards, that’s an hour more than a typical workweek (remember, these are actual working hours… I don’t get lunch breaks), but this is very difficult and draining work – not the sort of thing you can do for that long a period of time with no rest in between and emerge without exhausting yourself in some way.

It’s still not as bad as October 2006 – the week of my birthday was when I learned Matlab… forcefully, since I had to write code for six of our research projects in one week, and that was just for a workshop presentation. Some birthday that was πŸ™‚

Feyerabend

This man has earned my deep respect, for he was probably the only honest scientist remaining to the modern era. He alone disbelieved the self-delusions that most scientists still retain to this day… by subjecting the scientific method itself to scrutiny and recognizing the disparity between the process scientists claimed to follow – and the one they actually used.

His magnum opus is “Against Method”. It should probably be required reading for anyone seeking a scientific career.

Science

How did I become an anti-intellectual intellectual? πŸ™‚

After conversations with friends who share stories of collaborators not doing work, submitting fraudulent results, discarding reason, etc., I’m rapidly losing my confidence in modern academia altogether. I just want to finish my degree and do some real science already. It’s about ideas, not publications. It’s about advancing knowledge, not merely making the method sound like the solution to a grand problem. It’s about communicating results clearly and concisely, not obscuring them in pages of pseudo-intellectual jargon. It’s about openness, not skepticism. It’s about freedom, not censorship. It’s about learning, not servitude.

Until today’s academics realize this, almost all of science is a wasted endeavor. The few that keep it going in spite of the corruption are to be commended.

Dissertation – Week 11

As you may have noticed, it’s been 3 weeks since I last posted an update on my dissertation. I had submitted a draft of the first 50 pages of my dissertation and was waiting for a response. However, if I wait any longer for a response to this draft before proceeding, I am going to jeopardize my timeline. Therefore, I am going to resume work, first by reconciling the disparity between one of my formulas and the results returned by the tensor toolbox, then by performing my own experiments.

I just hope I’m not first told I need to revise everything when I submit the completed work and prepare to defend!

To be honest, I’m not sure how I’m supposed to write 100 pages of methodology on this topic. I’m still not sure I get the point. Our collaborators are supposed to be filling me in on this, but they seem to have disappeared; I haven’t heard from them in over a month.

I’m going to split the rambling that this is leading me down off into another post.

A Peer Review Benchmark

While discussing different strategies for revamping peer review in order to eliminate some of its many flaws, I came up with a benchmark to test any system’s false dismissal rate against. (Many consider peer review to have failed only if it accepts a paper that should have been rejected, but I consider the opposite a much more grave mistake).

A system is sufficient if it would have permitted Evariste Galois to publish his mathematical work. That’s it. Without changing any of his circumstances, including his general rejection by elite mathematicians (or government) of the time or the poor reputation of his academic institution, I am looking for a system that would have allowed him to circulate his papers (which would later prove revolutionary, after all) uninhibited.

Any system that fails this should be burdened under the knowledge that it would have denied us most of the field of abstract algebra and all ensuing discoveries – basically all of 20th century mathematics.

Using Physical Properties and Forces to Cluster?

It seems plausible to create clustering algorithms based on gravity and the Coulomb force, with masses or charges corresponding to specific point weights. A “cluster” then becomes the resulting “solar system”. For example, if we represented all objects in the solar system with their masses and distances, the theoretical model would label them as one cluster (“Sol”).

Another idea I’ve been toying with is to use the concept of physical momentum with gradient descent (I don’t believe this is the same thing as the existing technique called “gradient descent with momentum”), such that an “energy counter” is kept that increments when the gradient points downward (proportional to its magnitude) and decrements when it points upward. This will cause the optimization to “roll” down slopes, completely clearing small minima, which tend to be pathological. The result is wherever the optimization/rolling comes to a halt. (Nevermind, this is in fact the same thing, or almost so)

Of course, I still think estimating the minima of the MSE curve from what is already known of it then moving there to check would be much faster and possibly more accurate.

Finally, another idea is to deform a surface to minimize the local MSE of its k-nearest neighbors at each of several regions. I’m not sure if this replicates the behavior of an SVM with a kernel, however, but it should probably operate much more quickly than the cubic learning of an SVM hyperplane due to the local nature of the constraints.

Here's another example

Quoth the New York Times on both perfectly valid research being ignored and how a field becomes “hot”.

The usual pattern is taking place here: first, someone comes up with new results which the scientific community either ignores or rushes to discredit because they’ve somehow got the idea that skepticism (essentially making themselves unreceptive to ideas that don’t fit with their dogma) is the best way to evaluate scientific discoveries:

“But when those tools emerged in the early 1990s, Dr. Dick found stem cells in acute myelogenous leukemia, a blood cancer. He reported that such cells made up just 1 percent of the leukemia cells and that those were the only ones that could form tumors in mice.

Yet Dr. Dick’s research, Dr. Wicha said, β€œwas pretty much ignored.” Cancer researchers, he said, were not persuaded β€” and even if they had accepted the research β€” doubted that the results would hold for solid tumors, like those of the breast, colon, prostate or brain.”

Potential avenue that opens up all sorts of treatment possibilities presented, but they “weren’t persuaded”. Nice.

Now, wait for it…

“That changed in 1994, when Dr. Wicha and a colleague, Dr. Michael Clarke, who is now at Stanford, reported finding cancerous stem cells in breast cancer patients.

β€œThe paper hit me like a bombshell,” said Robert Weinberg, a professor of biology at M.I.T. and a leader in cancer research. β€œTo my mind, that is conceptually the most important paper in cancer over the past decade.””

Ah, so you “weren’t persuaded” when one person found it in AML, but if it’s in breasts, well, that’s a whole different story! Now it’s the most important paper in the past decade! After all, even if stem cells may not form in solid tumors, there must be no value in treating what is probably the most virulent form of leukemia, right? (I had described this exact sociological phenomenon of violent swings in opinion with new presentations just yesterday on a Slashdot discussion thread – also, why the disporportional emphasis on breast cancer? There are other cancers that kill many more people, have much higher fatality rates, and strike both sexes equally).

Now we see an idea becoming “hot”:

“Dr. Weinberg and others began pursuing the stem-cell hypothesis, and researchers now say they have found cancerous stem cells in cancers of the colon, head and neck, lung, prostate, brain, and pancreas.

Symposiums were held. Leading journals published paper after paper.”

Etc.

It’s a good thing they’re taking this theory seriously. It’s a bad thing that they ignored it for as long as they did because they were too convinced that they already knew everything to take a valuable hypothesis seriously. Sure, demand proof if you’d like. But don’t take absence of it or a perceived lack of quality as evidence that the hypothesis is wrong. Science is not law; the burden of proof can be taken up by others if they are not satisfied with the evidence because ultimately, we are all in this together.

Here’s another interesting tidbit from a linked article, which in my mind supports my theory that one tumor could supplant another if injected into the same site:

“They then injected laboratory-grown cancer cells into the benign tumors, which spread swiftly throughout the teratoma clusters. The result, they believe, is an ideal test bed for anticancer agents.”

This in turn supports my idea that it is possible to alter the characteristics of a tumor by harvesting and injecting particular cells. Selecting the weak in this manner can possibly make tumors more sensitive to treatment.

Still waiting on the equipment and training – or collaboration with someone who has such equipment and training – to actually test that one πŸ™‚

Dissertation – Week 8

A lot of stuff is going on this week, including a conference deadline for two papers, and because it’s still being reviewed, I don’t think I should continue working on my dissertation this week. Once it’s reviewed, I will resume.

I might still write 2 or 3 pages today if I find some free time, though.

More is less

I’m noticing that the more information you put into a paper, the more the reviewers demand (example: add an ROC curve to your new results in addition to class accuracies and suddenly they ask why you didn’t compute one for the old results that you’re citing as well!) This is probably why papers also tend to get needlessly long, which was the subject of another post some months back. Sort of stupid, but that’s what you get for getting people started on your ideas. The ironic thing is that the more a paper gets people to think about similar issues, the more successful it probably is – and yet because of this phenomenon, the less likely it is to be accepted!

If this hypothesis is correct, it would not only indicate that peer review causes rejection of perfectly good papers spontaneously, but that it actively seeks good papers to reject.

Motor learning rates

While testing my hypothesis on motor learning rates, I noticed that while there does appear to be a variance in the slope from person to person, some of the data appears not to make much sense. In particular, one person had a POSITIVE slope in his frontal lobe.

Now, what that essentially means is that this person had to do more and more processing with each repetition of the task. Tasks require less cognitive processing with each repetition; this is how we learn to do things like walk.

So what on earth is a positive slope supposed to signify? πŸ™‚