Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!


Forgot your password?

Comment Re:Education is getting better (Score 1) 205

| Well, you can argue that they "ruined" math education, but they weren't "idiots." The New Math was developed in the 1960s mostly by college professors and advanced math people in reaction to the "Space Race." The idea was to introduce mathematical abstractions (set theory, formalizations of analysis, etc.) at lower levels in education, which might be beneficial to students who were heading toward engineering and science degrees.

The Soviets had a better idea. Teach standard mathematics faster to the brightest students, and kick their ass.

Personal experience: Newly entering graduate students in physics who were from Russia said that they had covered the material in the mathematical methods for physics course in secondary school before college. That means, for example, single variable calculus at age 13, multi-variable at age 14, and by the end of high school, ordinary & partial differential equations, complex and real analysis, linear algebra, Fourier/Laplace transforms, and a bit of group theory.

Comment Re:This isn't AI.... (Score 1) 149

| This is a huge advance, even if it is only optimising multiple moderate-depth playing engines.

It isn't doing that.

They trained some strong neural networks to first predict move probabilities from 30 million expert moves and positions. That was just the start. Then they used 'reinforcement learning' where they played games against itself and propagated back the final outcome (game won/lost) all the way through the net space to improve the learning to the correct outcome (game won vs lost) vs matching expert's moves.

Then they used samples from those (huge sampling space), taking one single position from synthesized games, to train a one-step position evaluator deep network. Along the way they trained a simpler (1000 x faster) policy network to evaluate potential moves.

And then finally the actual algorithm mixed stochastic monte carlo simulations using the fast policy network (playing a subset of moves to game completion) with the trained deep value network. On a decent sized cluster with GPU acceleration.

Fixed moderate depth engines (as in chess) are a major generation behind even for conventional Go engines---there was a generational shift to the Monte Carlo deep tree searches some years ago, and the new method uses that approach partially, augmented by big deep learning networks.

It's a huge achievement, but it was also a huge amount of work.

I know nothing about Go, some stuff about machine learning, and I did download and read the full-text paper.

Comment Re:The Future! (Score 5, Informative) 149

I've read the paper.

It doesn't quite use a "brute-force" approach, but it certainly does use significant, and intelligently designed, Monte Carlo searches which are informed by well-trained neural networks. The neural-network alone approach, without any Monte Carlo search during play, is not as strong, though it does appear to equal a state of the art conventional Go program. See Figure 4b.

And the training of the neural networks and construction of their training sets certainly did need quite a bit of 'brute force' as well as 'efficiently wielded force in large quantity'.

Comment Re:There are two reasons for this (Score 1) 460

| Isn't it odd how as a percentage there are so few women, children and old people along with these 20 something male "refugees".

No, not at all. The men of that age are subject to conscription by any faction of torturing thugs, or summary execution on sight. And there is no employment other than the hell of a poorly paid, poorly lead, and expendable soldier.

Of course the men are going to go first rather than subject their wives and children to unexpected dangers---they will make money and then bring them over.

Comment Quite a bit of not quite true stuff in there (Score 2) 412

There is such a thing as boosted fission weapons, which do have fusion fuel---deuterium and tritium, in the core of the fission primary. This is not an "H-Bomb". The fusion fuel provides comparatively insignificant energy output from fusion and contributes almost nothing to the yield---however, it does provide an extra boost of neutrons at close to the moment of maximum criticality, therefore substantially increasing the efficiency of the fission reaction. It is a physical 'neutron gun', and in practice, a key step towards significantly smaller and lighter fission weapons suitable for a mass-constrained ballistic missile warhead.

The transition from fission weapons to true multi-stage radiation coupled thermonuclear weapons (Teller-Ulam) is indeed quite challenging scientifically, there are far more uncertainties than with the fission weapons. It's all about energy transfer, exotic thermodynamics and fluid mechanics.

There are still significant undisclosed secrets in this stage as well. The fusion section is not just Li6-D, but a combined assembly of fusion and fission fuel & tampers. A major part of yield in modern thermonuclear weapons is in fission of the secondary, and it is very incorrect to say that they are "clean weapons". A big part of yield (60-80%) is from fission and the amount of fallout is proportional to total fission events & energy.

A boosted primary core is a practical prerequisite for multi-stage H-bombs, though as it provides a cleaner and more appropriately shapable radiation pulse to drive the secondary.

I believe it to be more likely that DPRK tested a boosted fission primary and the staff told His Supercritical Eminence that it was a H-bomb. Which is true, from a certain point of view.

Comment Re: Private industry... (Score 1) 85

I have heard personally from a former colleague who is intimately involved with LANL, a high-level University researcher with ties to LANL but not a direct employee.

The current lab management contractors have various metrics for which they manage the lab employees & programs. One metric which is now completely absent? Progress, results, and success in innovative scientific research.

The UC management might have been lax in other ways---it was very hands-off and let the lab do anything it wanted as long as there was some opportunity for UC professors to also work with LANL.

But now, the fundamental purpose of much of the lab is not even recognized----and the management fee far, far higher. And previously when the UC was the prime contractor, all the money it got for management it put back in to joint research.

The national labs can do things that universities cannot---sustained research and in particular development that takes too long and would not be rewarded in a cutthroat academic environment, but the bean-counting compliance-oriented, instead of success-oriented, management philosophy is not appropriate for what is literally, as the name says, a National Laboratory.

LANL has always been the best, and I believe that some of its excellence has been because it was managed by the University of California primarily, and others were managed by private contractors.

Comment Re:Tried that in Vietnam too... (Score 1) 290

The targeting of precision guidance against ground targets, especially fixed ones, is much better now than in Vietnam.

Now against a modern air defense system, the cruise missiles and incoming JDAM's are susceptible to counter-fire as well, of course. Then the question becomes which side has enough.

The longest range SAM's are quite expensive.

Comment Re:Peak Aeroplane (Score 1) 290

| That's pretty much why airframers aren't putting to use hybrid wing body aircraft or other new concepts that have been modeled or developed in recent history.

For passenger planes, I learned there's a significant problem with pressurization. An oval tube is the easiest and lightest to keep pressurized, and to be manufactured that way. Blended wing designs are more difficult and require more structure contributing to more mass and lower performance.

Comment Re:Contested vs. uncontested sky (Score 1) 290

| If you had to ask me which aircraft I would want a squadron of for providing ground support, in a ground war against Russia or China? I'd definitely pick a squadron of A-10s over the F-35. They'd be more effective, nevermind that I'd probably have money left over for some F-16s on top of that.

Not at all. Modern missiles are so much better than before. Improvements are in sensors and software.

I used to think as you did---but in a permissive air space environment like Afghansitan, there's still the AC-130 which is being upgraded and preserved.

Comment words are A-OK (Score 3) 131

| A simple language is wordy, requires a lot of typing and keywording to express a thought. Cobol has things like "move a to b", which is simple and readable, but requires a lot of typing. Basic has things like "for index = 1 to 10 step 2" which is very readable, but requires you to type out "step" and "to".

I have learned how to type and I can type "step" and "to" rather quickly now, and certainly remember what they mean, a whole bunch faster than remembering what ^%::() means, right now in this file in this programming language given that I have to sort of remember 5-7 of them for my job.

I'm just an ordinary human with an ordinary human neocortex. Us in the hoi polloi have evolved brain circuits for natural language which have been trained to recognize sequences of letters known as words, and we've learned what they mean somewhat automatically since we were about 4 years old.

Note that when we have to explain "for index = 1 to 10 step 2" we don't say,. 'well of course it really means j/x&&*10(2)", but instead we use the communication protocol which has words in it. A cortex is not lex and yacc.

Slashdot Top Deals

10.0 times 0.1 is hardly ever 1.0.