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.
- 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."