Slashdot Mirror

Why not ask Marvin Minsky.
I learned a lot by reading his stuff.
I disagree with some of the limits he puts on things but he certainly has the behavioural
aspects categorized.

He probably knows some bright prospects.

It'd be nice to see Marvin's site slashdotted...

http://web.media.mit.edu/~minsky/

This is an area with lots of crackpots, but also lots of really interesting stuff.

How do you tell the good stuff from the crackpot?

The good ones are published in top machine learning, computer vision, robotics, and AI conferences and journal. The crackpot stuff doesn't survive peer review.

Here are a few good examples:
- Geff Hinton (U. Toronto): http://www.cs.toronto.edu/~hinton/
- Yoshua Bengio (U. Montreal: http://www.iro.umontreal.ca/~bengioy/
- Yann LeCun (NYU): http://www.cs.nyu.edu/~yann/index.html
- Andrew Ng (Stanford): http://ai.stanford.edu/~ang/
- Sebastian Seung (MIT): http://hebb.mit.edu/people/seung/
- David Lowe (U British Columbia): http://www.cs.ubc.ca/~lowe/

You've got to quit trying to advance on separate fronts. People have been exploring and reinventing the same old niches for sixty years. Little has changed except for the availability of powerful hardware with which to realize these disconnected bits and pieces. What is needed is a way to bring the many different segments of the AI and robotic communities together, because the solution is not to find the "winning approach", but to realize the value of the various perspectives and combine efforts. This is not a new idea, it is an old one which apparently just doesn't fit into the established research environments. Go to the library and read some old books on AI if you really want an appreciation of how pathetic the progress of ideas (not hardware) has been. To whet your appetite try some of Marvin Minsky's old papers - http://web.media.mit.edu/~minsky He recognized this situation nearly 40 years ago.

Enigmail is an OSS plugin for Thurderbird that gives GPG/PGP support.

Firefox and IE also don't have built-in Flash or Java support, but we all fix that within the first 5 minutes of an install, right? Email encryption should be no different.

The hardest problem I find is getting people to maintain their keys and a real trusted way to exchange keys w/o man-in-the-middle attacks.

Just putting your key in pgp.mit.edu or on your homepage doesn't prevent man-in-the-middle attacks any more than an SSL cert not signed by a CA that your browser already trusts is worth anything (again, unless you securely download that self-issued SSL key).

Haha, Scratch is awesome!

I wish I was still 8 so this would be impressive instead of an appalling waste of my time: http://scratch.mit.edu/projects/eurica/193902

Instead I have got him started with scratch which he loves. It's much better for introducing maths, logic and generic programing skills and it's a lot of fun.

He has done several homework projects in it which have been well received but I discovered recently that the teachers need to view his work outside of school because the local education authority firewall has a rule to actively block access to scratch ! I wonder if thay had a powerpoint presentation at couty hall with a slide labeled Scratch - Must stamp out.

Your argument in a nutshell is: "If we optimize a counter-strategy, they will optimize their own counter-strategy, so we are better off leaving the status quo favorable to their present strategy."

The status quo isn't favourable to any strategy - that's why it's a good idea. If you increase the probability that Arabs will be searched, the terrorists can gain an advantage by using non-Arab attackers. If you increase the probability that men will be searched, they can gain an advantage by using women. The only strategy that can't be exploited in this way is random sampling.

Should we stop associating bank robbers with people who walk into a bank with ski mask and gun on the premise that this will simply let robbers without ski masks and guns slip through undetected?

Masks and guns are necessary for robbers and unnecessary for non-robbers, whereas dark skin isn't necessary for blowing up a plane.

If you can isolate a subgroup that contains 85% of your "targets," it is simply logical that 85% of your resources should be dedicated to that particular subgroup.

No, that strategy can be exploited.

This has no similarity whatsoever to racism, which is the (contradicting) fallacy of replacing demographic weights with a general assumption of intrinsic character traits.

I never said it was racism, I said it was bad security.

And you are quite mistaken to suggest terrorists have so much facility in choosing the demographic to cull recruits.

They only need four or five people to carry out an attack. Do you really think they can't find four or five recruits in the whole world who don't look like stereotypical terrorists, i.e. young Middle Eastern or North African men? Islamism is an ideology, not a race.

At present, there is absolutely no doubt profiling is *efficacious*, although please do note that I am not assuming that just because it is efficacious that it is *morally right*. You can still make against profiling even if it's known to make the best use of the available resources.

I'm glad you made the distinction and I understand where you're coming from, but even in narrow resource-allocation terms, randomness is still the best strategy.

... the intelligent kids have fewer and fewer excuses with places like MIT offering their challenging courses for FREE - http://ocw.mit.edu/OcwWeb/web/home/home/index.htm

I'm getting tired of the "all the intelligent people are victims", what really needs to be done is to have good guidance counsellors and to know about these internet resources, many intelligent kids can get the help they need from professors on the net and whatnot now. They have all the ability, what they need most is to have a map to be pointed in the right direction.

Life, my eyesight, both of my legs, etc. Some things can't be priced b/c they are irreplaceable and vital to my happiness.

On the contrary, we can and do price those things every day. The price can be measured by studying what salary is demanded of people to bear what percent risk of loss. See "Changes in the Value of Life, 1940-1980", at http://web.mit.edu/costa/www/risk10.pdf, and "Present Value of Expected Lifetime Productivity, by Age, Gender, and Discount Rate, 1992", at http://www.nida.nih.gov/EconomicCosts/AppendixB_1.html. They crunched the numbers and built an equation showing the risk-versus-wage-demanded curve for average people.

A "100% risk of loss of life" is a special case where the price curve goes to infinity, because money loses its value at the time of death... although in the Middle East we do observe suicide bombers being incented by large (for them) financial benefits awarded to their families. And it's easy to imagine a parent selling (if it were possible) their remaining life in exchange for the money needed to cure a loved one of some terrible ailment.

It would also be possible to get the price for consensual loss of eyesight, if our society allowed people to sell their eyes as donor organs. You too would consider it if the price was right -- say, ten billion dollars cool cash. It's easy to sit in your armchair now and insist that "no price is too high", but it would be a different story if a truckful of cash actually pulled up in your driveway.

Of course, you should be suspicious of me anyway; I'm over 30 :-)

So am I. :P

There's an upcoming paper coming from MIT on this topic, Recognition of Natural Scenes from Global Properties: Seeing the Forest Without Representing the Trees that proves this isn't as hard as you might think.

To sum up this massive paper in a very small (and likely highly imprecise) nutshell, building models up from basic objects (the traditional method) is only one way to approach this. Using this method, you are correct; it's impossible to understand what a canyon is. Using the new global properties methods in this upcoming paper, you can gather basic elements that could easily help in assigning location properties, understanding that something pictured is a desert or forest, and theoretically using that data to help determine which desert or forest (this latter portion is beyond the scope of the paper, but great fodder for a future paper that builds upon these fundamentals).

While the method currently requires a high level of labeling in its images, it is hoped that this labeling becomes unnecessary on larger data sets.

You could try Trustworthy Systems by Bernstein. I own the book, but never actually read it because the lectures from Bernstein's associates at Stevens Institute of Technology did a sufficient job to explain the examples and concepts in the book.

Another favorite of mine is the work of Nancy Levenson and her students at MIT.

Just about every US state has a law that slower traffic must keep right. In California and Virginia it's this applies *regardless of speed limits*. Blocking the left lane because somebody appointed themselves speed limit enforcer is more dangerous than the act of speeding itself.

http://www.mit.edu/~jfc/right.html

MIT's OpenCourseWare uses Real for their video lectures. I can't think of any other website that does, but for me, this was more than enough reason to install the player.

The Apology by Plato is a good start. One of the themes in it is useful skepticism. "I know that I have no wisdom..." It's an easy short read. http://classics.mit.edu/Plato/apology.html

According to google.org Google has donated $33 milion from AdSense adverts to more than 850 nonprofit organisations in 10 countries throught the world.

The Make-A-Wish foundation has received more than 25% of all the online donations from Google.

Google has given more than 30% of all the yearly donations to the Doctors without borders program.

Google has also donated to the Grameen Foundation located in the US and $2 mils to the OLPC project...

As it was pointed out by another poster, no 1024-bit RSA is not sufficiently strong. Recent papers have demonstrated that factoring a 1024-bit key is now within practical reach. See for example this PhD dissertation from a student whose advisor was Shamir (the S in RSA FYI), which estimates that cracking a 1024-bit key would cost a few million US dollars.

Sure, at this point only a small number of organizations have a few million dollars to spare on cracking RSA, but this is beyond the point. The flaw is sufficiently serious that security standards are now recommending 2048-bit RSA keys minimum.

What I am talking about are relatively recent developments, it is not very well-known that 2048-bit is the minimum recommended length. This is why 1024-bit keys are still wildly used everywhere. My bank (www.wellsfargo.com) uses a 1024-bit key...

"a ready-made surveillance system to controlling 3rd world governments"

We've had that here for ages, why did it take them so long to catch up? If not then why have I had six separate visits from the 'anti-terrorist' police and why did the "BBC", come in and photograph all the staff.

Welcome to the desert of the real .. :)

It does not change the variable. In Python, a variable is a lexically scoped named reference to a data object. That means that (lexically - by position in the source code) the same variable can point to several different objects at the same time, if you create multiple closure environments (e. g., by returning an inner function from within the enclosing function, where the inner function refers to a variable in an enclosing scope), or simply if you have several frames of the same function on the call stack, each of them has "its own variable X". And conversely, you can refer to a single data object from multiple variables (which makes a difference undesr certain circumstances, see below). But that does not matter that much, the most important thing is that each of these is just a reference to something, and the qualities of this 'something' are of importance here.

The most important thing is this: The values (real data objects) in Python can be mutable. When you introduce mutability of data objects (and even the mutability of variables - those "mere references" - in case of languages with first-class closures) into a language, your equivalence predicate essentially has to split into two. Just a single '==' (or '=' in not-quite-like-C languages) is no longer sufficient in case of mutable data. In Python, you get an 'is' operator for the "most strict" identity of objects. If 'a is b' in Python, it is required that 'a == b' also be true, but the implication does not work the other way round. If you have a=[1]; b=[1]; c=b in Python, then (a==b, a==c, b==c) returns (True, True, True) and (a is b, a is c, b is c) returns (False, False, True).

One important consequence of 'a is b' being false is that mutating a does not influence b. The other consequence of introducing mutability is that it introduces the time factor into the system, essentially breaking things like "the order of evaluation does not matter" that work in "pure math". (You can see that in certain C expessions like c[i] += a[i] * b[i++] - the result is in fact undefined - please correct me if I am wrong. Here the variables themselves are the data objects, as they are values, not references, but that is just a detail.) But is that an "abstraction leak"? I am not sure. Perhaps you should study the Chapter 3 of the marvellous SICP book (I can not recommend it enough), named Modularity, Objects, and State, and decide on your own whether it is an "abstraction leak". Maybe from the point of view of a mathematician, but there are certain benefits that it brings. (See the linked text. And no, I simply cannot see Your Average Java Programmer grokking monads any time soon, or at least not all of them.)

The very fact that the data object referred from b changes when you change a actually proves here that b is "the same as" a in the most strict meaning of "sameness" that Python has to offer. But "sameness" is a relative concept and it very much varies from language to language, and very often you find yourself switching between different concepts of "sameness" even when working within the bounds of a single language, dependening on the issue at hand. And you have to rely on the actual specification - if your language has one. :-) Only in languages like Haskell are you operating with all values immutable.

I'm not really sure what Neanderthal modded this down, but I'm foregoing use of my mod points to say that you are dead-on. In the college scene, many of the leaders are men in technical majors (engineering, computer science, material science, etc.) and there is a perennial shortage of leaders. So yes, nerds *are* actually fought over by cute girls. And the dedicated girls have this tendency to want to know how to lead, so that they can teach their friends to dance as well.

Additionally, ballroom dancing is actually quite popular in nerdy schools. One of the largest dance competitions in the U.S. is held at MIT (yes, that MIT), and if our school could afford to go there, we would...

I'm surprised there's even one semiconductor company in Massachusetts (it's about as far away as you can get from the tech centers in the US)

Haskell should appeal to their sense of mathematical beauty. It is an efficient, terse, and pure language.

Another more mainstream but similar language is Scala.

Don't forget Structure and Interpretation of Computer Programs -- required freshman course at MIT. All course materials online.

By the way, a really fantastic introduction to programming for kids is Scratch.

Haskell should appeal to their sense of mathematical beauty. It is an efficient, terse, and pure language.

Another more mainstream but similar language is Scala.

Don't forget Structure and Interpretation of Computer Programs -- required freshman course at MIT. All course materials online.

By the way, a really fantastic introduction to programming for kids is Scratch.

Indeed, I taught a course based on SICM last year. As a mechanics textbook it is superb -- it clarifies many issues that are ambiguous or wrong in other books, and it gives a very nice treatment of nonlinear dynamics and phase-space structure. Its treatment of canonical transformations and canonical perturbation theory is the best I have seen. For those brought up on Goldstein, it is a godsend. Thank you, Sussman and Wisdom.

Now the caveats. For programming, it uses Scheme (like SICP) -- but, unlike SICP, the programs and libraries they require work only on MIT's dialect of Scheme. If you are a Scheme guru, I imagine you can teach a brilliant course on it. The advantage of Scheme over Fortran/C/other non-lisp languages is that you can combine symbolic manipulation and numerical work: you can derive Lagrange's or Hamilton's equations with your program, pretty-print them and then solve them numerically. The disadvantage is that you need to learn Scheme, and the farther you go the more non-intuitive it gets (since they never teach it formally, and claim their students at MIT pick it up on their own). In fact some programs as printed have bugs, fixed in the errata, and I really can't see the error. So after a while the programming part just became a glorified graphing system, useful to do phase space plots of a driven pendulum.

The book's approach to computer work is completely different from what most physicists think of as "numerical methods" (and they never actually explain how to solve a differential equation numerically: they use canned routines from their libraries for that.) Nevertheless it is a very interesting approach and I'd recommend checking it out. It may have something in common with how physicists use Mathematica, Maple etc, and will teach you more about programming than those tools will.

Purely from the physics point of view, as I said, the book is outstanding. It is available for free here so it costs you nothing to take a look.

The exact language isn't so important as is flow control, file handling, basic methods/technique, basic resource management, and troubleshooting.

A solid understanding of the nature of floating point numbers wouldn't hurt either. For example, something like David Goldberg's What Every Computer Scientist Should Know About Floating-Point Arithmetic.

As to the language not being important, I don't know that that's entirely true. Each language offers a choice of types and some sets of choices are more instructive than others. I personally think Lisp or Scheme are good teaching candidates because they offer arbitrary precision integers and rational numbers in addition to basic floating point number types so that it's easy to see side-by-side the trade-offs being made between correctness on one hand and space/speed on the other hand that go along with choices in this regard.

Plus, if you go the Scheme route, you get teaching materials focusing on good abstraction like Sussman and Wisdom's Structure and Interpretation of Classical Mechanics , the proper companion to Abelson and Sussman's popular CS text Structure and Interpretation of Computer Programs !

The exact language isn't so important as is flow control, file handling, basic methods/technique, basic resource management, and troubleshooting.

A solid understanding of the nature of floating point numbers wouldn't hurt either. For example, something like David Goldberg's What Every Computer Scientist Should Know About Floating-Point Arithmetic.

As to the language not being important, I don't know that that's entirely true. Each language offers a choice of types and some sets of choices are more instructive than others. I personally think Lisp or Scheme are good teaching candidates because they offer arbitrary precision integers and rational numbers in addition to basic floating point number types so that it's easy to see side-by-side the trade-offs being made between correctness on one hand and space/speed on the other hand that go along with choices in this regard.

Plus, if you go the Scheme route, you get teaching materials focusing on good abstraction like Sussman and Wisdom's Structure and Interpretation of Classical Mechanics , the proper companion to Abelson and Sussman's popular CS text Structure and Interpretation of Computer Programs !

NPEs will become less widespread when JSR 308 becomes a reality, mostly because of @NonNull. You can read about the proposition here.

Obama wants to stop the manned space program for five years and give the money to education.

on 2. On the Effectiveness of Aluminium Foil Helmets demonstrates that the paranoid should be even more fearful of Tinfoil hats.

Before this gets too out of hand, let's make sure that we're all dealing with the facts. A quick Google search shows that Daniel Weitzner is also a lawyer.

"Mr. Weitzner has a degree in law from Buffalo Law School, and a B.A. in Philosophy from Swarthmore College."

Granted, I'd still rather have him advising the president than a former media executive, but he's not exactly Robert Morris.

Did you just summarize Rider's thesis?

This is similar to Sam Madden's CarTel project at MIT.

their (presumably FOR SALE) software

Well, no. As Hays pointed out they have a really huge database of images, plus by limiting severely how many bits it takes to represent an image they've made it possible to do the search in real time, all in RAM, with a modest investment in computing power. In addition, it's not simply pixel matching. The images are arranged according to a semantic tree of English nouns so, if I'm interpreting it correctly, you will end up with images next to each other on the map that are also semantically similar. This is a big difference from other techniques. Still it does rely on Google, et. al. image searches in the first place to semantically label the images.

Again, thanks to Hays for this link: http://people.csail.mit.edu/torralba/tinyimages/.

Read the papers then
http://people.csail.mit.edu/torralba/tinyimages/

What you're asking for is ill-defined, but much sought after.

A reasonable descriptor which produces distances that seem somewhat correlated with human perception would indeed be Antonio Torralba and Aude Oliva's gist descriptor.

http://people.csail.mit.edu/torralba/code/spatialenvelope/

It's become quite popular in computer vision and computer graphics for scene matching.

The actual paper is at http://people.csail.mit.edu/torralba/publications/nipsRecognitionBySceneAlignment.pdf

From what I can tell, it's basically, "blur the image down to only a few hundred pixels and then you have less data to comb through!"

tin foil hats have been found to be a government conspiracy, don't you? They have exactly the opposite effect you desire.

See http://web.mit.edu/newsoffice/2001/ocw.html to read about how MIT answers your charges in explicit detail. They clearly state that what you're paying for is the interactive experience, the ability to ask and have questions answered, and, oh yeah, the degree. It's from 2001 so you're getting angry 7 years after the fact.

Personally, I think that if someone has the time, energy, and drive to learn something this way, more power to them.

I also believe that these lectures are less useful than a book would be in learning as they can't possibly go into the same depth. And just like I'm not going to complain about my professors putting a free book online, I don't care about the lectures.

I think you may have missed the altruistic mission that many universities have of encouraging learning in all forms, by all people - not just those sitting their butts in a local classroom.

I really hope that same type of open-source economic irrationality will help fund my open-source FPGA tools startup!

my point is that there are a dirth of FPGA boards with better cost/performance value that could be used to prototype a graphics rendering FPGA system. Physical hardware isn't the limiting factor to an open source graphics card; the open source FPGA 3-D rendering code is the real missing piece. In fact, making a board was probably a distraction for this project because by the time the firmware is ready for real graphics workloads the FPGA on-board will be obsolete.

Here's some examples of 3-D engines for FPGA from the 6.111 lab at MIT:
3-D Pong (using rasterization):
http://web.mit.edu/6.111/www/s2006/PROJECT/7/main.html

Ray Tracing:
http://web.mit.edu/6.111/www/s2007/PROJECTS/5/main.html

There are hundreds of videos and code for FPGA projects up at http://web.mit.edu/6.111 (see project appendices for code).

I really hope that same type of open-source economic irrationality will help fund my open-source FPGA tools startup!

my point is that there are a dirth of FPGA boards with better cost/performance value that could be used to prototype a graphics rendering FPGA system. Physical hardware isn't the limiting factor to an open source graphics card; the open source FPGA 3-D rendering code is the real missing piece. In fact, making a board was probably a distraction for this project because by the time the firmware is ready for real graphics workloads the FPGA on-board will be obsolete.

Here's some examples of 3-D engines for FPGA from the 6.111 lab at MIT:
3-D Pong (using rasterization):
http://web.mit.edu/6.111/www/s2006/PROJECT/7/main.html

Ray Tracing:
http://web.mit.edu/6.111/www/s2007/PROJECTS/5/main.html

There are hundreds of videos and code for FPGA projects up at http://web.mit.edu/6.111 (see project appendices for code).