On a lighter note...

Oct 07, 2008

My next 10 year research agenda

In one of the Christmas dinners, my host insisted that I tell about my research to the audience (all more than 70 years old). Now, after showing them the dishwasher unloading work and fetching an object demo, here is how the conversation went, and that pretty much helped me decide on my next ten year's research agenda:
A 75 year old lady: "So, how does it work?"
"Well, it learns to pick up objects. It was shown hundreds and thousands of images of objects, and when shown a new object it uses its learning to pick it up"
"So, it is matching a new image to its recordings of other images."
"Yes. Thats the idea yes." "It learns to identify visual ..." (trailing off)
"So, it is just matching what it sees to the recordings of thousands of images"
"Yes..."
"Oh. Then its not a big deal. My grand-daughter, when born, can pick up toys even when she has not seen anything. Your robot needs to record thousands of images"

We have a long way to go....

May 15, 2008

3d girls

From a Make3D user: "can you give me some tips on how I can edit/fix the photos to make the photos of the girls "3d"?"

Jan 28, 2008

Slashdotted

"I wonder if we can get a 3-D image of their server room on fire."
---someone on slashdot, Jan 28, 2008.

See Make3D.

Jan 24, 2008

Anything for research. STAIR 1.0 went out in rain --- anything for research. (See image.)

Aug 1, 2007

"If this paper were to be correct, we do not need any of the many depth estimation algorithms."
--- someone from planet Earth, where they still think digital watches are a pretty neat idea, July 29, 2007.

The field of computer vision can be characterized as immature and diverse [...] there is no standard formulation of "the computer vision problem". Also, and to an even larger extent, there is no standard formulation of how computer vision problems should be solved. Instead, there exists an abundance of methods for solving various well-defined computer vision tasks, where the methods often are very task specific and seldom can be generalized over a wide range of applications. [...] most practical computer vision applications, the computers are pre-programmed to solve a particular task
--- From Wikipedia, Aug 1, 2007

Vogons are one of the most unpleasent races in the galaxy. Not evil, but bad tempered, bureaucratic, officious and callous. They wouldn't even lift a finger to save their own grandmothers from the Ravenous Bugblatter Beast of Traal without orders signed in triplicate, sent in, sent back, queried, lost, found, subjected to public enquiry, lost again, and finally buried in soft peat for three months and recycled as firelighters.
--- Hitch Hiker's Guide to Galaxy, Timeless

"Systematic writing of a very significant portion of the paper in footnotes is, as the reviewers point out, clearly an on purpose circumvention of the submission guidelines. ...... As appropriate, some significant discussions were carried out among several senior members of the committee regarding this, as it is a policy question. In the end, all four senior members who were consulted consider this to be unacceptable. The content of the paper is very good and it is surprising that interesting research is not matched by good judgement on the part of the authors."
--- someone trying to bring his species remarkably close to the Vogon race, Jan 10, 2007

Feb 2, 2007

Dec 8, 2006

Astronomy is the study of stars and planetary objects. Since equations that govern multi-body problems cannot be solved analytically; scientists use simulations and approximations. You can hear statements like this in field of Astronomy: "At 14:33:40 on 12/14/2003 if an object is launched in such and such fashion, it will be able to pass through every planet in the solar system because of peculiar formation of the objects." For me, such statements are same as a future prediction made by a gypsy.

Astrology goes a step further. It is a field in which you go beyond predicting physical trajectories from arrangement of heavenly bodies, to predicting behaviour of organisms from positions of heavenly bodies. You can hear statements such as: "If you start your company at 10:02:59 on 1/31/2004, then it would be successful." Looks unscientific, doesn't it ? How can one find a correlation between position of heavenly objects to space-time trjectory of a organization ?

We all agree, there might be some correlation because of laws by Newton and Einstein --- even in presence of randomness introduced by this quantum stuff. But how is it possible for Astrologers to find a correlation or a functional mapping from position of planetary objects to the space-time trajectories of a multi-body system, such as an organization on Earth ?

If you were in the field of Machine Learning, you might buy this arguement: Yes, indeed, such a functional mapping is possible; and some people in distant past discovered this mapping. Here is how Machine Learning works: You have lots of data, you think of some model/parameters from your intuition: kernels, bayesian parameters; you learn those parameters from the data; and in the end you come up with some mathematical function which even the best scientists have no clue how it works. It just magically predicts output from the features! Now, I would argue, why astrological predictions could not be science developed in the past, for which we have forgotten the principles and are left with just the inference rule?

Oct 11, 2006

I was recently reading stories by Isaac Asimov. (Yes, again after a long time.) I looked at them in a different perspective --- no longer do they seem mysterious (as they did in 1996 to me), no longer do they seem funny (as they did in 2000). Rather, they seem to "real" stories of the present times, i.e. 2006. It surprising that you can read all about facebook, orkut, google scholar, microsoft word spell checker, computer-based stock trading, in his essays written 50 years back or more!

Mar 31, 2006

I completed my undergraduate degree in Electrical Engineering at IIT Kanpur (India), completing some projects in which I put a lot of man-hours programming the computer to do something, like recognize facial expressions, speech, or design circuits. Now, at Stanford for my PhD thesis, it would imply that I have to put more and more man-hours in programming.

The trick, a professor in AI told me, is to ask the computer to write the thesis! Therefore last night, I asked my computer to write my thesis titled "Forty-two" -- and the computer crashed for some reason. I got the lesson -- these days I am just asking it to learn to use its arms -- thinking that once it learns to use them, I will ask it to write my PhD thesis, while I can muse on some more interesting problems.