How hard is it to get a patent? What about a patent that applies Neural Networks to a problem?
Do you need to file a provisional patent before getting trying to get funding for your idea?
Do you have to build out a model first if you are trying to patent neural networks or can you get funding to build the model and change the patent as you learn more and go?
Big corps will sidestep it and chinks will just rip you off anyways. Why bother with it in the first place?
Patents are worthless if you can't afford the legal fees to defend it in court.
Trolls may buy your patent and wait until some high rank bureaucrat wants to punish a corporation.
Is there a market for patents? Like a browseable one?
Like a market for stuff you can buy? Not as far as I'm aware. But there are services like Google Patent and others similar that let you review public patents, and you could probably scrape academic directories to find lists of patents produced by active academics. Those are more likely to be unused, since academics may patent services while waiting for someone to come and buy them, or to get prestige points.
>to get prestige points.
How do they let others know the patents they have? How much are they willing to pay for prestige? Maybe there is a market for it...
Is that a decent idea? Become a patent troll enabler? Maybe you could make up ideas that target corporations that are currently woke but are liable to be bought out like Twitter so that the market on them goes up when the CEO does a wrongthink
patents are worthless, anyone with half a brain and know english can file a patent. you just need to hire lawyer to use the correct language. most patents are useless piece of garbage. neural networks ideas are obsolete after 3 months anyway. if your idea is good enough, publish a paper on it. otherwise, don't even bother.
>if your idea is good enough, publish a paper on it.
Does publishing a paper give you any rights beyond the 1 year protection any type of public publishing gives?
>neural networks ideas are obsolete after 3 months anyway
That's the thing, I need funding to build the model and determine the actual path the design will take. Right now I'm looking at a litany of options. My idea will bring in a huge economic benefit to my community and many others worldwide, but I want to take my cut, it's helping people make more money literally.
>My idea will bring in a huge economic benefit to my community and many others worldwide
i'm serious, I feel like governments should be paying me to design and implement it
Just rent Google cloud or Amazon cloud to train your model then
The issue is collecting the data isn't going to be as easy as doing it inside a lab. I need grants to buy some land and equipment with to then send that data to some cloud service.
The thing already exists its just so funny because it's an overlooked thing but there is a tremendous need for it beyond the obvious market
>My idea will bring in a huge economic benefit to my community
Thirdworlder detected. Indian, I'd say
>Can't think in terms of uplifting his community
Atomized and racialized Merrishart noooticed
>Patents are worthless
>But publishing a paper, which will expose your IP to the world with no protection, not even claims of trade secret, and that you have to pay money for, now that's the way to go!
Academicucks, I swear.
you don't know shit anon. most ideas in ML are worthless in a few months. instead of spending a few thousand dollar to file a patent on your useless idea, publishing it to a top tier venue might get you a comfy, high paying research job at a tech company instead. it's the best way to get the most out of your useless idea. I've filed two patents in my grad school and I know how this bullshit system work.
>How hard is it to get a patent?
Harder than drama queens claim. About 50 percent of patent applications are granted.
>What about a patent that applies Neural Networks to a problem?
Quick check chows 147,872 hits, more come every day. Roughly 30000 new applications annually.
>Do you need to file a provisional patent before getting trying to get funding for your idea?
You don't need to but the filing strategy is something you should discuss with a patent attorney.
>Do you have to build out a model first
No need for that.
>if you are trying to patent neural networks or can you get funding to build the model and change the patent as you learn more and go?
You can file more applications as you go along and discover more. This too is part of a patent strategy. It is not cheap, so you need funding.
>patents are worthless
Samsung, Google, IBM, Huawai and Microsoft have each more than 1000 patent applications for neural networks. You think they would have spent all this money if there was no value?
>neural networks ideas are obsolete after 3 months anyway.
Citation needed.
>if your idea is good enough, publish a paper on it.
So that others can steal your idea? Got it.
have you ever read the ever fucking shit out of these patents? they're old ideas regurgitated (aka 5 years old or older) with a slight modification to bypass the patent requirement so that they can avoid getting sued by competitors, nothing new.
>Citation needed.
if you ever worked in this fucking field, you should know this.
>So that others can steal your idea? Got it.
No one want to steal your stupid idea, chud. You are not the people behind Transformer.
case in point:
https://patents.google.com/patent/US9406017B2/en
>muh patent are so innovative.
people who actually filed and submitted patents know they're fucking worthless except in commercial settings. and the only reason is just to avoid being sued by some random ass patent hacking company.
>have you ever read the ever fucking shit out of these patents?
Some, yes.
>they're old ideas regurgitated (aka 5 years old or older) with a slight modification to bypass the patent requirement
Normally you have to fulfill th epatenting requirements to get an application granted. Just how do you get a patent by bypassing the requirements??
>so that they can avoid getting sued by competitors, nothing new.
What?
needed.
>if you ever worked in this fucking field, you should know this.
OK, so you have zero source, got it.
>>So that others can steal your idea? Got it.
>No one want to steal your stupid idea, chud. You are not the people behind Transformer.
So your source was your sigmoid, got it.
>people who actually filed and submitted patents
You don't file patents, you file patent applications.
>know they're fucking worthless except in commercial settings.
This makes no sense. Few file applications for fun, it is mostly about the business at hand.
>and the only reason is just to avoid being sued
Only? Unlikely. You could use tech from expired patents (See Gilette defence) or preemptively publish before anyone file an application.
>by some random ass patent hacking company.
What on earth is this?
Just get it anyway, unless it's really big and you will need lengthy legal proceedings to protect your IP then I would advise going to a trustworthy corporation that potentially needs it (preferably after the patent)
Did some patent work as part of an incubator.
>How hard is it to get a patent? What about a patent that applies Neural Networks to a problem?
You need to prove that the thing you're making is doable, that it's meaningfully different from other equivalent patents or licensed software, and the patent office will want proof to test and review. It's also harder to patent things that you've publicly disclosed in certain venues. E.g., if you publish a research paper describing the details of your neural network then you have to disclose that to the Patent Office and it will limit the scope of what you can patent. Also, what's patented about it can't be something common. E.g., you can't patent "neural networks that use X Algorithm" if X is a publicly known algorithm.
>Do you need to file a provisional patent before getting trying to get funding for your idea?
It's a good idea, but not required. I would recommend doing so if you intend to get a patent at all.
>Do you have to build out a model first if you are trying to patent neural networks or can you get funding to build the model and change the patent as you learn more and go?
All you need to get funding is to have something people want to invest in. Having a patent can be useful, but it has to be meaningfully different from what's done elsewhere. You can't just patent neural networks, you'd have to patent a particular instantiation of neural networks for a specific purpose, and that may not provide meaningful value to people who are investing in you. It's something you should actively discuss with potential investors, in particular because you may be able to secure funding to allow you to hire a patent lawyer.
Does this same advice apply if you are seeking grants, namely to fund research and patent the idea. I read about small business grants that seem to apply to the domain I'm doing.
>you can't patent "neural networks that use X Algorithm"
Lets say that I'm using neural networks to solve a problem that was already solved using something that isn't a NN, but my method is unique in that it can use less inputs among other reasons, and the benefits could be pretty large using my methods, but I'm not sure exactly which pipeline I'm going to use until I start using data I collect (which will cost me money to do, maybe I could self fund but I'd be limited, why not apply for a grant instead if I can?). I'm assuming that I'd need to have a provisional patent application in at least before I do right? Then, when I get funding, can I update the patent as I go when I realize the correct ML pipeline I want? Isn't it such that your provisional patent doesn't actually get disclosed?
For the majority of software, like the code you write when creating your neural network, you don't file a patent. Software typically comes with an implicit copyright which is based on the license you include with your source code. MIT and GPL are popular open source software licences. There's other licences you can use that are more restrictive. The most restrictive licence is no licence at all, which is usually just called unlicensed, which means people can look at your code but they can't copy it. If you're using an existing library to create your neural network then you yourself are probably beholden to various licences, for open source it's usually that if you modify that library code in your own code then you have to share the modified library. But what seems to be common in AI is the developer will release information about the code and maybe their model but not the source code. Facebook recently created an AI that played poker and was able to consistently win and they told everyone about it but didn't release the code or the model. So basically if you want people to not copy your code then don't share it online
I wonder how specific patents that use AI or ML techniques have to be other than "the neural network"
You would be patenting an industrial process where ML involvement is one step. Sort of like how chemical production has patented processes but the reactions aren't patented.
>Sort of like how chemical production has patented processes but the reactions aren't patented.
This is awesome news, and I'm assuming the process plants use improve with iteration and the company doesn't have to file patent after patent update due to them not needing to include it in the patent?
I'm not sure, maybe a patent is required for some higher level process our methodology but for source code you pretty much never have a patent, you have a copyright
yeah mine is essentially solving a higher level problem and ML is just one part of a pipeline
Yeah I think a patent is more for an invention whereas copyright is for things you write etc. When you write some code it's not usually considered an invention on it's own. But like you said the overarching process of what the code does and connects to etc I guess could be an invention
Has anyone ever tried to get a small business grant for research?
better to just form a secret society cult and deify yourself with all the technologies you produce.
rope high society by charging ludicrious amounts and just kill anyone who tries to leak the technology.
Kek! The best part about this type of business is that I get more customer data my models get more and more "magical" so in a way there will be no competition with me once i get enough momentum
doesnt alot of code and digital products fall under copyright protections more than patent laws?
like microsoft isnt patented software is it?
its a copyrighted work under trademark.
So how can the patent office test your patent if you don't give them the software to test? For a patent with a NN can't they just logically figure out if it will work or not?
>How hard is it to get a patent?
very easy.
It's only worth it if you have the proper operation set up to actually make money on it, viz. big companies
https://www.tinaja.com/glib/casagpat.pdf
>https://www.tinaja.com/glib/casagpat.pdf
Did you even read this hot garbage? It starts off with
>(B) Shooting a sequence of kiddy porno videos in the basement of your favorite church.
and then considers
>(B) will both have considerable upside potential
It starts by hitting rock bottom, and then it proceeds to dig even further down.
In general getting a patent isn't too hard, but doing so for Neural Networks is pointless because of how fast the field changes.