1

Let us suppose there is a patent with the claim: "a method for performing machine learning technique on a device." How to interpret this term machine learning technique? There are many machine learning operations and some of them are using matrix multiplication internally. Does it mean that performing matrix multiplication on the device independendlty from machine learning process is also patented or it has to be in context of a machine learning process to be patented?

Specifically, I am actually referring with the question to the first claim of the US7219085B2 patent.

4
  • The quoted text does not appear in the cited patent.
    – Eric S
    Apr 9 at 15:12
  • Yes, it does not, but I am also interested in understanding meaning of that particular phrase (machine learning technique) in any context
    – Marko
    Apr 11 at 6:23
  • I created an answer for you. Normally technology questions are off topic, but I know a little about this subject.
    – Eric S
    Apr 11 at 12:13
  • Thank you for the answer. Well, I wouldn't call this question technological because I am not interested in technology as is rather that using operations that the technology is also using. I think you can see that from the last question in a comment I made.
    – Marko
    Apr 11 at 13:33

2 Answers 2

2

You’re quoted words "a method for performing machine learning technique on a device" might be a paraphrase from the abstract but do not actually appear in the patent. In understanding what a patent covers, words matter. The words that matter the most are the claims.

  1. A computer-implemented method comprising: processing a machine learning technique using a graphics processing unit to obtain results, wherein processing a machine learning technique further comprises using a pixel shader to compute an inner product that is at least one of: (a) a vector inner product; (b) a matrix inner product; decomposing the inner product into sub-problems and performing multiple passes over the sub-problems using pixel shaders; using the results to provide solutions for use by a computer application.

If you are asking if to infringe the claim you need to be doing the steps (a) and (b) in a certain context, the context would need to be processing a machine learning technique with a pixel shader.

The scope of "machine learning technique" would be interpreted as one skilled in the art would understand it further boxed in by the specification and the prosecution history. That means that if, to overcome a rejection, the applicant stated a limited meaning of machine learning technique then they are stuck with that limitation when it comes to infringement.

The other independent claim 14 is much more specific about the machine learning context in that it specifies "training learnable parameters."

3
  • Actually, the exact quoted text doesn't appear in the linked patent at all.
    – Eric S
    Apr 9 at 15:13
  • I edited the answer, thanks.
    – George White
    Apr 9 at 16:15
  • Thank you for the feedback!
    – Marko
    Apr 11 at 6:26
1

I see from your comment you are most interested in what a machine learning technique consitutes. You could do worse than simply reading the Wikipedia article. I'm not an expert, but have dabbled in the implementation of machine learning in a field which I am an expert in.

In a nutshell, machine learning is a class of computer algorithms which use large amounts of example data to train a model which can then be used to predict a result from unknown data. For example you might want your model to be able to tell you if image contains a balloon. You train your model on perhaps 10,000 images which have been prejudged as to whether they have a balloon. Then test the model on another set of images where you know the answer. If all goes well, the model can then be used to detect balloons in images with high accuracy.

The training process for machine learning algorithms such as convolutional neural networks is very numerically intensive. They can be parallelized however which means that you can split the computation to run an many processors at once. Graphical Processing Units (GPRs) are comprised of large numbers of processors so can be used to speed up training dramatically. This particular patent is an example of one such method.

6
  • Thank you for the explanation. My concern is the interpretation of the machine learning technique. For example, if I want to run some program (e.g. matrix multiplication) on a GPU, will be I liable for infringement of the patent because matrix multiplication operation is commonly used in machine learning operations and techniques (such as neural networks and convolutions)? Or, maybe better formulated, will I infringe the patent if I use operations that machine learning operations are made of but not directly related to learning and other processes important for a machine learning model?
    – Marko
    Apr 11 at 13:27
  • @Marko The cited patent has not patented matrix multiplication using GPUs. IT hasn't even patented using GPUs with machine learning in general. The patent uses specifically pixel shaders. What I can say is look at the claims. If you are doing each and every step of a claim, then you infringe on the patent. Matrix multiplication is not machine learning. There is no learning going on.
    – Eric S
    Apr 11 at 21:48
  • @Marko I should also point out that there are widely used open source machine learning libraries available such as TensorFlow which work on GPUs. Nvidia, the largest maker of GPUs even teaches you how to do matrix multiplication on their GPUs for free. docs.nvidia.com/deeplearning/performance/…
    – Eric S
    Apr 11 at 21:56
  • Thank you again! I understood what you have said. Here is one more question for you. Wouldn't some function that is utilizing a GPU for computing (e.g. NVIDIA API) be considered equivalent to a pixel shader under doctrine of equivalents because they are doing the same thing: take input and calculate output using a GPU? I am curious about your opinion on that.
    – Marko
    Apr 13 at 11:18
  • @Marko I’m not a lawyer. I believe people have been using GPUs to multiply matrices for well before the cited patent.
    – Eric S
    Apr 13 at 13:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .