leecarraher

powerpc/IBM is still a big player in the server/HP computing market. They do many cool things with their architectures since cost is less of a factor(dynamic smp, switcheable endieness, OMI) but they suck to build code for from an out-of-box experience standpoint.

I agree, this is a classic do as I say not as I do situation. Punish businesses for paying out ransomware ransomes meanwhile the government(mainly local, but sometimes larger) pay out routinely when their systems are compromised.

Instead of focusing on other countries successes with HSR it may be more applicable to look at our ongoing attempts at high speed rail. The California HSRA has been working toward the goal of HSR in california for over a decade. California has the largest tax revenue in the US and one of the most progressive populations. Despite this, they still are plagued by cost overruns, deadlines, and political opposition.

"A serious Man" is "fully based on the concept of Schrodinger's Cat" , what? It is a post-modern retelling of The Book of Job, not sure other than the main character being a physics professor. At one point a failing student states that he understands Schrodinger's Cat, to which he is immediately admonished, for the analogy being pop-physics, and without understanding the math it is useless.

oh then this is just approximating the values of the linear transform. I figured at least they'd use mlp to learn each layer of the FFT butterfly network, and the corresponding weights. What would be of interest would be if you could extend this for n length fft using an RNN network

Alt title, can a neural network learn the fast fourier transform?

What doesn't defeat AI just makes the GAN smarter

I don't see anything in the plan about reviving the Yucca Mountain Nuclear Waste Repository. This is a key storage facility that's been used as a political scapegoat for far too long. No one lives near the facility, it's set in one of the lowest seismically active areas in the US, has almost no possibility of leaching waste material into groundwater and the environment. Yet people will fight tooth-and-nail against it for what, to preserve dry cask storage in people's actual backyards, in populated areas, with undetermined contamination risks.

This is a pretty clickbaity title in my opinion. Bubble sort in lower level language X is going to be slower than quick sort in high level language Y. And bubblesort in high level language is going to be faster than merge sort for small data sets on low level language X. If you aren't comparing the same underlying routine, or data application, I don't think any conclusion should be made. Comparisons between languages is exactly why asymptomatic analysis was devised. Extract the process from the low level and hardware characteristics, and get the overall complexity growth. But this doesn't work the other way around. You can't compare different routines in different languages and expect big oh to be comparable.

Join your local unions if possible: https://en.wikipedia.org/wiki/The_Animation_Guild,_IATSE_Loc... , and support union made motion pictures.

Smaller is probably better for evs. In this class of vehicles evs stand to get the full benefit of a considerably lighter and less complex drivetrain, that ice cannot come close to. Furthermore marketing them as short range, low speed people movers means the overall battery weight, which is likely the biggest impediment to ev replacing ices, can be better compensated for.

Yes, not pointing out the difference between coding some novel technique and a well defined software project, completely misses the reason the code is often not well organized. Suggesting that researchers are bad programmers is just a lazy excuse, somewhat damaging, and by no means the rule. I wrote a large complex framework for my research and the very nature of it causes me to add modules and techniques for parts I didn't know would work. And at times hard forks for when I wanted to try something new, which merging back would be impossible to do cleanly. At times you have a hunch and like a fever dream, change who knows what, but you just have to see something through. There is no waterfall method, kanban and agile makes no sense here and even unit tests are I'll defined.

The us has a similar profit driven model parallel to state funded schools. While private schools have always been available, only in the last few decades did they move from large monolithic institutions to national corporations. What we tend to call charter schools, profit off of government funding while sometimes putting corporate incentives over student well being. As with public schools, there is a spectrum of excellent to failing schools. But instead of the students success determining profit, the school is often faced with the decision: chase test scores to get more government funding, or file countless, often borderline false IEPs for difficult students and profit off the extra "need based" government funding. This had the effect of replacing the natural spectrum of education with a pigeon hole, excellent high stress test driven school. Or failing, school whose goal is more that of a prison than educational institution.

It is the primary training method for Pytorch and Tensorflow models and thus essential for training artificial neural nets (such as AlexNet, weird they would randomly pick an architecture). It is often used for non-negative matrix factorization, non-linear regression, the basis for most modern machine learning algorithms.

Stochastic Gradient Descent as well.

Awesome, just watched this two nights ago. fun to think even in 1992 it was still acceptable to show someone cracking a password by whittling down the passwords one character at a time. To their credit at least it was a bit better than war games, where it just locks in one character at a time. Where this one could maybe construe that it started with some hash of the pin stored on the machine and accessible via the card reader in reverse..., and they were maybe factoring it. Still an awesome scene and by far my favorite james cameron movie.

I suspect we will not see a revolution in deep learning from GNN without a corresponding hardware(FPGA?) or optimization method(HSIC? HTMs?) advance.

GNNs trained by backprop are making many of the same mistakes that LSTMs did: solve one important problem(exploding/vanishing gradients), but introduce a bunch of hyperparameters that bring along their own set of problems. GRUs are successful in my opinion, because they remove some of those tunable parameters.

Of course as the post suggested, not being able to tune something, often loses out against some more tuned and curated solution. GNNs being the newest version, having tons of parameters to tweek. In the end do we get a better, ultimately more generalizable solution? Or do we just get more hyperparameters to tune, and spend more time and money for a modest, unrepeatable gain.

The breadbasket as the name implies, grows mostly cereal crops: wheat, corn, barely and not a cereal but diverse use: soy. There are small operations that do grow more traditional vegetables, but due to their labor intensive harvests and produces short shelf lives, they are not terribly profitable (without underpaid migrant workers located near large populations to consume or process the produce quickly). This in my opinion is why we see things like water intensive almonds, only being harvested in places like water scarce California. Labor cost is something vertical farms often treat as a given, which at least for most of our non-cereal produce production, it is not.

I think this may comes down to interpreted vs compiled. Compiled language assumes some hardware instruction that interacts with primitives at some point. Where interpreted languages are using subroutines to deal with every object. So why not have a check if the value is in bounds, and if not, use arbitrary precision.

It seems like they've made a battery that only ballasts heat, with a material that is tricky to move around and has a complex mechanical cycle. It will be interesting to see what the eroi is on this vs. lithium ion, pumped storage, or even old school boiler heat ballast.

Cool hack, however i agree with the poster, since it appears to only have a 500KHz usable bandwidth(less than rtl-sdr) at an albeit faster (5x) sample rate, i probably won't have much use for it. Definitely a great reverse engineer and discovery journey though.

this is the new solar panels on everything, 3d print everything, water from air, spectrophotometer is a tricorder... wave of kickstarter-esque ideas that already have relatively well optimized solutions that they will pick on the drawbacks of while admitting none of their own. In this case, it's the maintenance nightmare something like this would be as an underground project with a falling risk.

Nice, well there you go.

Final Jeopardy : Trebeck: You know what, how about you just write down a number, any number at all. Could be a 1 or a 2, perhaps 3... and excel you answered; A smiley face emoji, simply stunning.

Excel is no longer motivated by the original intention of a spreadsheet, and now caters to the lowest common denominator, a piece of graph paper. As such MS has shifted focus from doing calculation to text and graphics layout tool. white text copied from a terminal : white text white background you got it! comma separated numbers : default a long string with commas in it want a plot : it is in the insert menu for some reason, since plots and numbers are no longer excels raison d'être

I like the idea, and have always been displeased by the lame underpowered inverters they often put in autos. That said, a compact, quiet 2.5kw inverting generator that weighs about 50lbs will only set you back about $300. Hopefully the addon is comparably priced.

At least that's Microsoft's motto

A better method would be to have reservation only items, or even just a sign up sheet and distribution. As the concept of people following hot tips and traveling to en mass to various locations around the city will almost certainly exacerbate the spread. If the majority of people are good enough to use the system for it's intended purpose, then they can maybe also just not hoard items others may need.

2 and 3 are unlikely, most in store inventory and ordering systems are quite antiquated, not easily integrating with modern web technology, they also do not monitor things like shrink and shelf misplacement. Working with a large grocer, the best method for in stock is sales analysis.

i use c in scientific computing for small processor intensive routines that can leverage openmp's straightforward parallelism.

My understanding was that quantum volume was an attempt to unify the previously used metric of number of qubits with some concept of potentially being able to do useful work. The last part being why these metrics are elusive. Adding more Transistors to a processor is only useful if we agree they are all beinh used. Same with qubits, in particular if you can't error correct your qubits you can't be confident in the result. So improvement in quantum volume seems to give a better yardstick for the roadmap on general quantum computing.