4 ms·

How are methods like these patentable when the research in these domains has been openly published and built upon by the community already? For example, the "Ge

How are methods like these patentable when the research in these domains has been openly published and built upon by the community already? For example, the "Generating audio using NNs" patent abstract is: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating an output sequence of audio data that comprises a respective audio sample at each of a plurality of time steps. [...] ... which seems really broad.

JoshTriplett 8y ago
Not necessarily; the major claim in the original WaveNet research was that they were the first to have a neural network directly generate audio samples. (Not arguing that these aren't extremely broad and problematic patents, but Google may well be the first in many of these areas.)
- BillyWM 8y ago
  How does an NN generate audio without generating samples? The outputs are somehow interpreted as a continuous function / polynomial with tons of terms? And nobody ever once tried to not do that?
CobrastanJorji 8y ago
Patents are a weird beast. They are written to be as broad as human language could possibly be coaxed to allow. The abstract matters for very little. There will be a list of specific claims later in the doc. These count for everything. The claims will usually begin obscenely broadly, practically "I have invented a way to perform calculations with electricity." Each subsequent claim will add an additional refinement/limitation to the previous one, and then lawyers will play games trying to see how broad of a claim can be admitted. The patent attorneys will eventually let the applicants claim the broadest claim for which no prior art can be easily discovered. Filing a software patent is a weird undertaking. You describe to a lawyer your idea, which takes like 2 minutes. Then they have you brainstorm for every possible extension of it you can think of for an hour or so. Then the lawyer disappears and comes back with a 60 page document that seemingly makes almost no mention of your actual idea. Then you sign that you definitely invented whatever the word salad says.
jsnell 8y ago
The abstract doesn't really have all that much bearing on the actual contents of the patent. Remember the lessons of "How to read a patent in 60 seconds" [0]. Skip the title, skip the abstract, skip the dependent claims. Read just the independent claims. [0] http://www.danshapiro.com/blog/2010/09/how-to-read-a-patent-in-60-second/ http://www.danshapiro.com/blog/2010/09/how-to-read-a-patent-...
BillyWM 8y ago
"when the research in these domains has been openly published and built upon by the community already" Well, mega-corps like Google lobbied for a first-to-file system (America Invents), for one thing. Community shmunity; whoever has the most resources to file the most applications wins.
- webmaven 8y ago
  FTFY: > whoever has the most resources to file the most applications soonest wins.

JoshTriplett 8y ago

Not necessarily; the major claim in the original WaveNet research was that they were the first to have a neural network directly generate audio samples. (Not arguing that these aren't extremely broad and problematic patents, but Google may well be the first in many of these areas.)

BillyWM 8y ago
How does an NN generate audio without generating samples? The outputs are somehow interpreted as a continuous function / polynomial with tons of terms? And nobody ever once tried to not do that?

CobrastanJorji 8y ago

Patents are a weird beast. They are written to be as broad as human language could possibly be coaxed to allow. The abstract matters for very little. There will be a list of specific claims later in the doc. These count for everything. The claims will usually begin obscenely broadly, practically "I have invented a way to perform calculations with electricity." Each subsequent claim will add an additional refinement/limitation to the previous one, and then lawyers will play games trying to see how broad of a claim can be admitted. The patent attorneys will eventually let the applicants claim the broadest claim for which no prior art can be easily discovered. Filing a software patent is a weird undertaking. You describe to a lawyer your idea, which takes like 2 minutes. Then they have you brainstorm for every possible extension of it you can think of for an hour or so. Then the lawyer disappears and comes back with a 60 page document that seemingly makes almost no mention of your actual idea. Then you sign that you definitely invented whatever the word salad says.

jsnell 8y ago

The abstract doesn't really have all that much bearing on the actual contents of the patent. Remember the lessons of "How to read a patent in 60 seconds" [0]. Skip the title, skip the abstract, skip the dependent claims. Read just the independent claims. [0] http://www.danshapiro.com/blog/2010/09/how-to-read-a-patent-in-60-second/ http://www.danshapiro.com/blog/2010/09/how-to-read-a-patent-...

BillyWM 8y ago

"when the research in these domains has been openly published and built upon by the community already" Well, mega-corps like Google lobbied for a first-to-file system (America Invents), for one thing. Community shmunity; whoever has the most resources to file the most applications wins.

webmaven 8y ago
FTFY: > whoever has the most resources to file the most applications soonest wins.