Often a patent will include background information to assist the reader in understanding the claimed invention. The text will usually also point out the shortcomings of prior inventions and how the present invention is an improvement. Neither of those are part of the invention being claimed.

To understand the scope of what is claimed, you always need to look at the claims. The claims will always specifically point out the scope of the invention. Note that all elements in a claim are necessary. If an accused infringer's activity is missing an element, then it does not infringe.

I think that if you look at the claims of the patent you reference, you'll find that the claims are limited to particular applications of sequence processing using particular neural network configurations.

Not sure about the patentibility of those patents (they are still all in the application phase, and I have not read them), but to my understanding patents a) have to be tested before court and b) often fall back to less generic claims in practice (claim no. N, with N > 1), if not being completely invalidated. Now, I expect Google to have good patent lawyers, but time will show the value of those patents.

Usually when patent applications are made they are written up to be as general as possible and will then steadily be whittled down until they're accepted.

That said, I really don't see how a number of these will get even close to validation. Quite a few are describing an absolutely generic, everyday process with so much prior art it's insane (recurrent neural network?). If I'm missing anything here I'd love if someone could correct and inform me.

Many times companies will file patents on any IP they design that is broad enough to disrupt their market. With the patent being exceptionally broad, it ironically leaves less room for questions when the patent request is filed. One would hope everyone on the Government side of patents would be well versed enough in the given technology but unfortunately that is rarely the case so having a broad patent is sometimes easier to push through.

This could be the point. Hopefully, they will be denied and we can all rest easy. The best defensive patent is something being declared unpatentable.

You are reading it wrong. The things they are patenting look like the following:

1. A neural network system implemented by one or more computers, wherein the neural network system is configured to generate an output sequence of audio data that comprises a respective audio sample at each of a plurality of time steps, and wherein the neural network system comprises:

- a convolutional subnetwork comprising one or more audio-processing convolutional neural network layers, wherein the convolutional subnetwork is configured to, for each of the plurality of time steps:

  - receive a current sequence of audio data that comprises the respective audio sample at each time step that precedes the time step in the output sequence, and

  - process the current sequence of audio data to generate an alternative representation for the time step; and

- an output layer, wherein the output layer is configured to, for each of the plurality of time steps:

  - receive the alternative representation for the time step, and

  - process the alternative representation for the time step to generate an output that defines a score distribution over a plurality of possible audio samples for the time step.

This appears to me to be describing an autoencoder for audio compression.

There are two other claims as well, all related specifically to audio processing.

Now whether those actual claims are valid is a separate question. I don't know the state of the art in 2016, they could have been. They are still math, I agree. But that's how you tell what they are actually claiming, you can entirely skip the body of the patent and just read the numbered claims.

The ones that don't start with "The ... of Claim ..., wherein" are independent claims meaning that they stand on their own. The ones that do read that way are typically used to narrow the independent claim so that when they say the independent claim is too broad they can make those dependent claims mandatory.

So for instance, if the USPTO (probably) rightly claims that generating audio with a convolutional neural network isn't patentable, they can fall back to saying "The neural network system of claim 1, wherein the audio-processing convolutional neural network layers include one or more dilated convolutional neural network layers." is valid combined with claim 1 because no one has thought of that specifically, and perhaps that's all that is granted.

In that case, as long as you don't use a dilated convolutional neural network layer then it is not infringing.

It's still all ridiculous nonsense, but that's how you read the patent.