Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
DEAL: For $25 - Add A Second Phone Number To Your Smartphone for life! Use promo code SLASHDOT25. Also, Slashdot's Facebook page has a chat bot now. Message it for stories and more. Check out the new SourceForge HTML5 Internet speed test! ×

Comment Re:FP16 isn't even meant for computation (Score 1) 55

So, one problem is that there is not always more data. In my field, we have a surplus of some sorts of data, but other data requires hundreds of thousands of hours of human input, and we only have so much of that to go around. Processing all of that is easy enough, getting more is not.

Also, by "effective", I should have made it clear that I meant "an effective overall solution to the problem", which includes all costs of training a wider, lower-precision network. This includes input data collection, storage and processing, all of the custom software to handle this odd floating point format, including FP16-specific test code and documentation, run time server costs and latency, any increased risks introduced by using code paths in training and , etc.

I'm not saying that I don't believe it's possible, I've just seen absolutely no evidence that this is a significant win in most or even a sizable fraction of cases, or that it represents a "best practice" in the field. Our own experiments have shown a severe degradation in performance when using these nets w/out a complete retraining, the software engineering costs will be nontrivial, and much of the hardware we are forced to run on does not even support this functionality.

As an analog, when we use integer based nets and switch between 16-bit and 8-bit integers, we see an unacceptable level of degradation, even though there is a modest speedup and we can use slightly larger neural nets. I'm very wary of anything with a mantissa much smaller than 16 bits for that reason--those few bits seem to make a significant difference, at least for what we're doing. We're solving a very difficult constrained optimization problem using markov chains in real time, and if the observational features are lower fidelity, the optimization search will run out of time to explore the search space effectively before the result is returned to the rest of the system. It's possible that the sensitivity of our optimization algorithm to input quality is the issue here, not the fundamental usefulness of FP16, but I'm still quite skeptical. If this were a "slam dunk", I'd expect to see it move through the literature in a wave like the Restricted Boltzmann Machine did.

Oh, and thank you for the like (great reading) and the thoughtful reply. Not always easy to find on niche topics online.

Comment Re:Exploitative by design? (Score 1) 153

It seems like these systems are exploitative by design, even if exploitation wasn't explicitly the goal. They're designed with every possible algorithm and available data to maximize labor output at the lowest possible cost. Individual workers are operating at extreme information asymmetry and against a system which does not negotiate and only offers a take it or leave it choice.

This is by far the best comment I've ever seen regarding this sort of algorithmic labor management.

Normally I'm all for this sort of thing--my company is a client and uses it to handle large bursts of data processing quickly--but the information symmetry argument is a powerful one. Also, there doesn't seem to be a lot of competition in this space, which might otherwise ameliorate a lot of the problems induced by the "take it or leave it" bargaining approach.

The analysis provided by the article is absurd, but yours seems to lead to the inescapable conclusion that some kind of regulation is necessary to prevent blatant exploitation. Maybe just reducing information asymmetry in some way, or requiring transparency in reports available to the public on the website regarding effective wages paid to workers as a fraction of the minimum and average wages of employees in their respective countries. Surely someone can find an answer to this.

Comment Reddit's management staff (Score 1) 410

Reddit's management structure is toxic, heavy-handed, and one-sided. And I say this as a proud leftist.
Authoritarianism is not exclusive to the right wing. Left wingers can be equally authoritarian.
And a lot of the time, we forget this. We are so proud of our ideological purity that we squash those who disagree with us.
And as a direct result, they rejected us in the elections. Because we stopped being liberals and started being the Fairness Police.

Comment Re:FP16 isn't even meant for computation (Score 1) 55

Accidentally posted as anonymous coward, reposting under my actual name.

So they're all excited about the lowest-precision, smallest-size floating point math in IEEE 754?

FP16 is good enough for neural nets. Do you really think the output voltage of a biological neurons has 32 bits of precision and range? For any given speed, FP16 allows you to run NNs that are wider and deeper, and/or to use bigger datasets That is way more important than the precision of individual operations.

There's a lot of rounding error with FP16. The neural networks I use are 16-bit integers, which work much, much better, at least for the work I'm doing. Also, do you have a good citation that FP16 neural networks are, overall, more effective than FP32 networks, as you've described?

Comment Terrible idea (Score 1) 159

Terrestrial radio uses frequency division AND geographic separation in order to provide communications ability to the users on this planet. To tinker with the ionosphere would increase propagation even for those signals for which propagation beyond line of sight was not a significant concern. Not every user needs over-the-horizon signaling. To a very large degree, we use the horizon effect for useful things- for one, it allows us to have multiple 100KW television transmitters all using the same channel all on the same continent. Ionospheric changes will have negligible effect on the power needed to adequately cover a metropolitan area with signal, and may actually INCREASE the power needed to swamp a distant signal that's intruding spatially into another market.

Comment The real problem (Score 1) 1145

I think the real problem is that there's a minimum required to not be homeless in most developed nations. There are ZERO apartments in most areas that would be described as "basic", having just a bedroom and a bathroom and a kitchenette. We are therefore burdened with having to have middle-class income to even live within the law. Nobody builds basic, poverty-level housing anywhere. This means that while you work in a gas station or restaurant, or wherever you're underpaid or not employed within your primary career, you're shit out of luck. This affects college students whose parents do not contribute, or can't, and disproportionately affects teens who grew up in state care as orphans or wards of the state. In those circumstances, when you turn 18, you're turned out onto the streets. If you make it, fine. If you don't, nobody cares. Having a brother in law who grew up in state care and who is currently homeless, I am always surprised at the sheer number of homeless people who have similar circumstances, and cannot get out of it because to maintain a job, you have to have a place to live, and to have a place to live, you have to have maintained a job long enough for first and last month's rent, a security deposit, and you have to miraculously have a decent credit rating, somehow, after being homeless. It's one of those things where there's a crack in society, and once you're in it, unless a miracle happens and someone offers you a place to stay for basically free, you'll never get out of it on your own.

Comment Re:It's Heartbreaking you're not in Jail (Score 1) 482

Several emails sent through the server state in absolute terms that the reason to operate a private email server is to avoid Freedom Of Information Act compliance and other Federal requirements and data requests. The entire purpose of that server was to avoid legal compliance and transparency. The reason she had her own server rather than a Gmail or Hotmail account is that Google or Microsoft could be subpoena'd and would comply and turn over records, but you can wipe or destroy a server that's completely under your control and utterly avoid recordkeeping for criminal or incriminating correspondence.

Comment Re: The Taste must have been fired also (Score 2) 474

The Gros Michel banana is not extinct. It just is rarely grown because it's vulnerable to a blight, and growing an entire field of it practically guarantees that you'll lose your crop. It is grown in smaller crops to guard against this. You can get them, but not cheaply, and not at your store.

Comment Re:How much was this advertisement? (Score 1) 99

It's actually NVIDIA that doesn't have something in the 200 dollar price point that's even remotely new. It's completely inappropriate to compare a 200 dollar card to a 325 to 350 dollar card. Most consumers don't buy 400 to 800 dollar cards. AMD knows VR cannot succeed without a mid-range mainstream card that is capable of VR, and without capability, people won't experience VR, and if they don't experience it, they won't demand better hardware, and AMD won't have business. But- If both vendors make VR capable cards at the 200 dollar price point, then there WILL be demand for VR, and after that, demand for even better hardware to run VR with, and then both AMD and NVIDIA make money. Nobody demanded video games on computers until computers with video games capabilities hit an accessible price point, and VR is no different. It is NVIDIA that decided to launch their 700 dollar card and 450 dollar card first, and delay the MAINSTREAM card. AMD instead launched the mainstream card FIRST, with the 400 dollar cards coming later.

Comment Re:Hold Ma Beer and Watch This! (Score 1) 1023

The argument that companies would rather have nobody rather than pay more is ludicrous and is NEVER USED with any other commodity such as fuel or flour. It exists purely to allow fiscal hegemonies to keep a vendor (us) from raising their prices. If your employer could make do without you RIGHT NOW, you'd be out the door. But there are certain times when you need someone's help in order for you to make money. And that's not going to change. There will always be times when the owner of a business needs a human being to do something for him. And what we have to decide on is whether it's gonna be that guy who pays the worker, or whether it's going to be the taxpayer.

Because starving potential thieves existing in large numbers is simply not an option, never has been. If 20 percent of your population falls into danger of starvation, your wealthy will be murdered and your government overthrown. No amount of Apache helicopters can fix a nation that has that kind of problem, and automation WILL bring us the potential for that level of problem. All that's required is a tiny dip in ambition among the investor class at this point.

Someone's going to be feeding the working people of this nation. I propose that it should be the employer.

Comment I doubt that they've contemplated the implications (Score 1) 209

First, Stone fails to imagine the possibility that his exposure to the workings of the NSA was a sanitized charade. Once the data is collected and stored, there is no limit to how many groups can make use of the data in its raw form.

Second, Stone fails to grasp the potential for such data. With this data, an incumbent and his cabinet can identify potential troublemakers, sort them by level of influence, and then simply selectively watch them until they break some obscure law, and prosecute it aggressively. Or even accuse them of having child pornography, since we already know that juries, the press and the public lose objectivity the minute the subject is raised, and the burden of proof goes right out the window. It allows the State to form an Enemies List, to know who dislikes what policy and how influential they are on social media or within their circle of activists.

From now on, we'll never be totally sure, when a politician or activist is prosecuted for something, that he was caught through the normal happenstance of law enforcement or whether he was expressly targeted because he opposes an incumbent or moneyed interests.

Comment Re:And the scientific evidence for this conclusion (Score 1) 391

> First, there is no reason to believe that we can built robots that can reproduce themselves.

What? This is exactly the technology humans are trying to reach! We're already a significant way down this path!!

> Second, there is no evidence that we or anyone else can build intelligent machines, as the original story seems to presuppose.

Nature did it. We can do it.

> Third, biological organisms are so many orders of magnitude more efficient and flexible than machines that it barely makes sense to put them into the same qualitative category "form of life".

This whole conversation is about extrapolating on the cosmic scale. If you look at the path robotics has taken in the last century it does, as pointed out, actually support the premise of this article.

> Hint: A human consumes only about 2.9 kilowatt hours per day, the equivalent of 1-2 light bulbs ...

Not relevant. Once machines are replicating and repairing themselves they'll do exactly what we do and find other sources of energy.

Frankly I agree with you that it's hard to picture Transformers inhabiting the universe, but OP did make a really good point that extrapolation isn't even in the ballpark of refuting this clown. Honestly I'm shocked he didn't come back with that XKCD cartoon.

Slashdot Top Deals

I just need enough to tide me over until I need more. -- Bill Hoest

Working...