Slashdot Mirror
Google Accused of Trying To Patent Public Domain Technology (bleepingcomputer.com)
An anonymous reader shares a report: A Polish academic is accusing Google of trying to patent technology he invented and that he purposely released into the public domain so companies like Google couldn't trap it inside restrictive licenses. The technology's name is Asymmetric Numeral Systems (ANS), a family of entropy coding methods that Polish assistant professor Jarosaw (Jarek) Duda developed in the early 2000s, and which is now hot tech at companies like Apple, Google, and Facebook, mostly because it can improve data compression from 3 to 30 times. Duda says that Google is now trying to register a patent that includes most of the ANS basic principles. Ironically, most of the technology described in the patent, Duda said he explained to Google engineers in a Google Groups discussion from 2014. The researcher already filed a complaint, to which WIPO ISA responded by calling out Google for not coming up with "an inventive contribution over the prior art, because it is no more than a straightforward application of known coding algorithms." A Google spokesperson refused to comment, and the mystery remains surrounding Google's decision to patent something that's in the public domain since 2014.
That's exactly what patents are for: To ensure huge corporations like Google keep tight hold on any new (or not so new) technologies.
It almost makes no difference whether Prof Duda can prove he invented this, if Google have a mind to they have the money to keep any case he might bring in court for as long as they like.br. The system is set up that way.
Patent trolls must die!!
This could also be a case where the employee, seeking to get rewarded for a patent, hid the prior art from Google.
"National Security is the chief cause of national insecurity." - Celine's First Law
The dumb shits in the patent office will sleepwalk through the rubber stamping process as usual.
Sounds like he didn't actually register a patent, but simply declared that his idea was public domain. The article isn't clear on exactly how he did so. The patent office won't necessarily count that as prior art, unless it's formally published. To actually prevent a company from monopolizing the idea, the most effective strategy would be to actually patent it and put it under a copyleft patent license... that is, patentleft. Of course the trouble with this is that patents are expensive.
This space intentionally left blank
So why not? Teh G RULEZ! Your World!
I'm pretty sure that's Google's motto.
-- Tigger warning: This post may contain tiggers! --
Even if you want to just give it away, you better apply for that patent or this kind of thing is going to happen Skippy... Now, if you really want to defend this, it's going to cost you a pile of $$, just to give it away.
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
At large companies, there are:
1) large incentives (several $K) to engineers to get a patent.
2) Incentives to managers to have their teams to get patents
3) Billable hours for outside patent council to file regardless of how questionable value that patent may be
4) Pressure from the board to CxOs have more patents than IBM
5) Easily fungible value as very, very few patents are ever licensed.
Big companies are not an individual who decides to file ANS patents.
Big companies provide incentives for their employees to file patents and employees submit patent proposals to some review committee and the committee decides which ones to apply for, then a lawyer works with the employee to write the application.
This whole process can happen without anyone involved knowing what the patent status of ANS is. In particular, big companies ask their employees not to search patents, because that creates triple damages risk where a plaintiff can show you 'knew' you were infringing because for example you downloaded the patent onto your computer and that was shown in discovery.
The underlying reason for this behavior is the triple damages provisions for willful infringement.
I should use this sig to advertise my book ISBN-13 : 978-1501515132.
And I don't mean countersuits in response to another company or individual suing them for patent infringement. I mean, has Google ever found someone violating one of their patents, and been the first to file an infringement lawsuit?
There's plenty not to like and to worry about with Google. But by my recollection, acting like a patent troll isn't one of them. If Google is trying to get a patent for an already-existing invention, then it's far more likely they applied for it just in case the USPTO was stupid enough to grant it. That way they would have the patent instead of some patent troll who could then sue Google for it. $10,000 for a patent application is cheap insurance against a patent lawsuit which could cost $millions to defend against.
... and investigate why the patent was filled. In my opinion.
Sounds like he didn't actually register a patent,
No indeed, he didn't.
but simply declared that his idea was public domain. The article isn't clear on exactly how he did so.
Wut ? It's right there even in the TFS on /. : he published the stuff back in the 2000s.
The patent office won't necessarily count that as prior art, unless it's formally published.
If you google a bit around :
- arXiv:0710.3861 - "Optimal encoding on discrete lattice with translational invariant constrains using statistical algorithms" first published in 2007 (that's the bat-shit crazy stuff that only a few mathematicians managed to understand but lay ground for the whole stuff)
- arXiv:0902.0271 - "Asymmetric numeral systems" first published in 2009 (second paper, where he re-visited these concept, and which spawned, among other the FSE - Finite State Entropy - implementation of tANS that is used by Yann Collet's Zstd - recently moved to facebook).
- arXiv:1311.2540 - "Asymmetric numeral systems: entropy coding combining speed of Huffman coding with compression rate of arithmetic coding" first published in 2013 and cites actual implementation such as ycollet's fse.
All these papers (which also cite actual real-world implementations) all predate Google's patent filing.
To actually prevent a company from monopolizing the idea, the most effective strategy would be to actually patent it and put it under a copyleft patent license... that is, patentleft.
Or you know, just publish it.
Like everybody else does in the academic world.
Formal publication DOES COUNT as prior art in most sane parts of the world.
That's also why Range-Coding, the predecessor of tANS and cousin of the patented arithmetic coding isn't patented itself: it was published.
Disclaimer: I've worked on entropy encoders for the compression of genomic data as part of the PoSeNoGap project.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Also, releasing something into the public domain means abandoning all rights to it. So rather than ensuring Google can't patent it,
He did so by publishing it (look my other post with arxiv refs).
These publications constitute prior art.
Google CANNOT patent it be cause by now, 2017, this techniques have been known for 10 years.
(Including successful implementation by Yann Collet's FSE and another one by one of the coders of the Farbrausch demo team).
he ensured that he has no standing to sue.
He can technically challenge the patent on ground of prior art.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Even if you want to just give it away, you better apply for that patent
Or you know.
Just publish it formally.
Like he did.
That constitutes prior art and make Google's patent invalid.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Google is now evil.
Bitch.
No, ANS improves processing performance by 3-30x. It provides the same compression ratio of Arithmetic Coding (the previous industry standard -- the 1977 IBM patent having long expired) but at nearly the same processing speed as Huffman Coding (from 1951).
Just to nitpick :
tANS (table Assymetric Numeral Systems) such as the FSE (Finite State Entropy) implementation by Yann Collet, at the hearth of Zstd compressor (now Facebook's) are NOT 3-30x better than other modern post-Huffman entropy encoder, such as binary-arithmetic encoding or range-encoding.
They are much *FASTER*. By lots.
They all boil down to the same logic:
do not use a fixed code-word book like Huffman (which is thus limited to integer number of bits).
but try to get as close as Shanon's theory predict the necessary bits, by subdividing number space.
range encoding works by dividing an arbitrary big number (usually an extremely long binary number).
For each symbol, you split this range in sub-ranges. More frequent symbol get a wider sub-range, more rarer symbol get a narrower one.
You pick the sub-range corresponding to the next symbol in the text you need to encode.
Then you keep the same work by subdividing *THAT* sub-range by the probabilities of symbol occuring in the position after that.
- Encoding relies a lot on multiplications (there isn't such a thing as a direct multiplication in transistors. Instead you implement it in microcode by combining shifting and adding. modern CPU have big shift-adders units, so they can manage in chunks of 64bits - on modern Intel it's a few cycles delay, 1 more if you want the upper 64bits of a 128bits product).
- Decoding relies on division (rules of three to "zoom" into the subranges), and division are fucking slow (again no such things as "division" with transistors. Instead you implement it in microcode, and most modern CPU tend to do it bit-by-bit meaning it's fucking slow. - on modern Intel it's dozens of cycles delay).
This thing was published in 1979. (And thus isn't patentable)
It's provable that it approaches arbitrarily close the Shanon limit. (all the symbols in the stream occupy a total of bits that is close the invert log2 of their respective frequencies).
Arithmetic Coding is a cousin technique patented more or less in the same era.
In can be seen as a special sub-case of range-encoding where the range is [0;1] and you subdivide it in fraction.
It's most often implemented as binary arithmetic encoding.
Input symbol are converted into a bit stream.
Then for each bit, the fraction is divided in two half. e.g.: if each bit has a 50:50 chance, the [0;1] is subdivided at 0.5
You chose the range under or above this mid-point depending if the bit is 0 or 1 and move forward to the next bit and sudivide the fraction based on the next probability (e.g.: mid point at 0.25).
Given that there are only 2 symbols, and therefore only a mid-point fraction to keep track of, its implementation is a little bit simpler to follow.
But because this works on *every single bit* of the input stream, you can guess it's either slow (on CPU - hence CABAC versus CAVLC in H264 videos) or require high-frequency on hardware implementation.
As it is basically a variation of Range Encoding it can achieve similar arbitrarily close to optimum encoding.
Then come Mr. Duda with his ANS.
They are basically range encoding turned on its head - to understand you must basically look at range encoding's bit the other way around.
- Range encoding works by subdividing an arbitrarily big number. ANS work by build a progressively bigger number.
- Range encoder usually works by writing out the most significant (high bits) away and then shift the working values. - ANS works by considering the least significant bit (the low part).
And here comes the real magic :
- With range-entropy: if you have 2 symbols with 50:50 chances, you subdivide the range in 2 halfs. With a different probability you still end up with 2 adjacent sub-ranges of different lenght.
- With ANS, because you figuratively "reverse the bits" : if you have 2 symbols with 50:50 chance, you'll find them alternating in
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
According to his LinkedIn profile he is no longer with Google https://www.linkedin.com/in/al...
What would be the Weissman score improvement?
.......busy doing evil........
Who said that?
Please remind me.
I forget who it was.
The guy who steals your TV should be way behind in who gets a prison cell and these folks front of the line.
i am a reasearcher from a private university from brazil and Google has been spying me and my coleagues for quite a while. but we are way, way smarter than them...
To hold a lawsuit for criminal conspiracy to commit patent fraud over the Goog's head as leverage to not fight the various fines that are being leveled against it.
Overrated
Moron moderator hates the truth about their Saudi buddies... Those bastards are fucking murderers! Why would you stand up for them?! It's like Trump defending the neo-nazis... There are no good neo-nazis, and there are no good Saudi motherfuckers. We should be glazing them over with our nukes more the the norks!
Doing the Wrong thing since 2015.
USA has changed the law and now it is not first to invent, but it is first to file who gets the patent. So even if he can prove he invented it, he can not stop Google from patenting it. But when Google tries to enforce the patent, others can cite the prior art to get the patent invalidated. So the assistant prof has to wait for Google to sue someone, then he can step in and help the defendant with prior art.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
The parent comment: " First, be Evil. I'm pretty sure that's Google's motto."
Google is now often acting in a way that is, in some ways, bad for users. What underlies that? Very poor management.
My guess is that Sergey Brin and Larry Page became overloaded running Google, now Alphabet Inc. Consider, for example, how you would feel about running a company with 72,053 employees. Extremely overloaded?
Now Pichai Sundararajan, also known as Sundar Pichai, is the CEO of Google. He was originally from Tamil Nadu, India. About 88% of Tamils are Hindus.
It is my understanding, based on talking with Hindus in India, that Hindus generally don't handle conflict well. For example, consider this quote from the article, Conflict resolution and Hinduism:
"The main teachings of Hinduism, according to Rao, include the following components:
An emphasis on suffering, impermanence and contingency;
Every act is suffused with spirituality and is meant to enable one to realize oneself;
Self-effacement, not self-assertion, is taught;
..."
To manage a technology company well, there must be deep emphasis on logical resolution of both technological and social conflicts. Conflicts must be resolved, not avoided. The quote above lists methods of avoiding the deep details of conflicts, not resolving them.
I'm guessing, and it is only a guess based on observation from outside the companies, that Google CEO Sundar Pichai is good at seeing methods Google can use to make more money, but is not good at understanding and resolving negative issues concerning those methods.
Now, technically knowledgeable people are complaining about many ways Google is abusive. For example, Google sells several services to companies that have web sites. The negative result is that Google tracks users of those sites.
A long time ago, I installed the Google Chrome Browser. The installation installed 3 system services. I uninstalled Chrome. I would not let Google's browser software to have more control over my computer than I have when I'm operating as a limited user.
See the article by Richard Stallman, The JavaScript Trap. Google is especially abusive: "Google Docs tries to download into your machine a JavaScript program which measures half a megabyte, in a compacted form that we could call Obfuscript because it has no comments and hardly any whitespace, and the method names are one letter long. The source code of a program is the preferred form for modifying it; the compacted code is not source code, and the real source code of this program is not available to the user."
The present emphasis of Google is apparently only on making more money. Alphabet and Google top management aren't paying attention to the social damage that is being done to the companies.
There are plenty of ways for Alphabet and Google to make money without having a negative affect on the companies, the users, and the world. Top managers who recognize and resolve conflicts can make healthy decisions. Top managers who think merely about making more money, and don't think deeply, damage their companies.
Companies like Google don't file patents as if it's some singular entity. Most, if not all, patents are submitted by individual employees. The attempts to file then go to the lawyers who look for prior art. But shit, they're lawyers, and although they're pretty fucking good at their jobs, they're not technical experts.
So occasionally, some asshole engineer decides to skirt the rules and take advantage of a lawyer who doesn't know the hottest new tech. And you get this news.
It's clear that this Professor Jarosaw Duda is a racist bigot with nazi views, and should have no voice in this matter. Google is on the right side of history.
Which you now have to fight Google in court to prove if they got the patent that you should have applied for.
*if* they get the patent...
Which exactly why Jarek Duda is writing to the patent office to inform them of the prior art.
Which is exactly what happened afterward.
Citing TFS on /. :
The researcher already filed a complaint, to which WIPO ISA responded by calling out Google for not coming up with "an inventive contribution over the prior art, because it is no more than a straightforward application of known coding algorithms.
Prior Art won.
Thanks to publications of Jarek Duda (see: arxiv), and provable discussion with Google (i.e.: disclosure) that predate the patent filing.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
The problem is that patents are supposed to be foreclosed by prior art, and publishing code someplace should forever make it impossible for a paten t to be issued. However, the patent office does not search, nor is the applicant required to search, for prior art, and the patent office seems to consider only prior patents when internally searching. I've had a few such cases appear when trolls sued someone using patents of stuff I published years back (and making code public domain was for a long time considered the sensible thing to do to keep it free).
The patent office should be much more careful, but the problem would be better solved by requiring that patent applicants search for prior art. At present an applicant need only say "I don't know any prior art" and if he is the Nth inventor of the idea (where N >> 1) he has no need to have sought prior art. Also if he has seen it but forgotten that he did, an applicant may believe he knows no prior art. That can happen easily and often innocently.
A requirement for search, which would have to show evidence of search not only of patents, but of public domain sources and open source information (magazines, journals, computer user group libraries, logs of old bulletin board systems, net repositories), could eliminate the problem and make it much harder for junk patents to be pursued. The person wanting to block all others from using an idea should have to bear the cost of this. In fact a retroactive requirement of this kind would be worth imposing, just to clean up such things. There are already requirements for paying maintenance fees for patents to keep them in force. It is probably getting easier to find some old public material - you would not need, say, a complete collection of the old Usenix tapes to find some evidence of what was on them. However, old magazines and the like are still publications. Some of the searching can be hard. For example someone claiming a variant of a B tree search (illustration, not serious real example!) should look also at programs in public whose source code might contain their variant, but which might appear as a DBMS system, or as some system that embeds a DBMS.
The cake is a lie.
"By the way if anyone here is in advertising or marketing... kill yourself." -- Bill Hicks
the designer describes the whole thing to the engineers at Google without an NDA in place. I'm pretty sure Google ends up with the patent in the end.
Sorry, I don't bother to follow TV/web series enough to get the exact reference (I stop at recognizing the "middle-out compression" part).
I suspect it's about an episode show the typical IP blunder of revealing some not-yet-protected company secret.
In the real world though, the NDA is completely superfluous in this situation.
The designer (Duda) here is describing something for which he has published articles nearly a decade before.
NDA - Non Disclosure Agreement - are exactly what it says on the tin : pacts asking the other party not to reveal any of the secret information.
That would be hard to do when there are no secrets because the whole thing has been in public knowledge for a long time.
That would be like asking for an NDA before discussion the wheel or fire.
And then pointing fingers at the guy who forgot the NDA, instead of point fingers to Google for trying to patent an already known thing.
Which is exactly what Duda is currently doing : pointing fingers at Google for trying to patent "as new" something which in fact has been published 10 years ago.
And Duda won : according to the summary the patent WAS NOT granted due to not being original.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]