Does anyone know is training a neural network on a GPU or multiple GPUs patented using Tensorflow library?
I want to know if using the library to train a model on a GPU or on multiple GPUs might infringe someone's patent, like this one EP3129870B1.
3
-
Can you please clarify? Tensorflow is an open source library. Anyone can use it.– Eric SCommented Feb 5, 2022 at 16:40
-
Please convert question over to: which keywords you have tried searching in which online database that then resulted in perplexing results returned. Asking for •opinion• on subject matter (i.e., inventory of all prior patents related to GPUs & TensorFlow library and/or its underlying) for which they have little prior expertise. Conversely if you show attempts to search using FreePatentsOnline or Google Patent Search or other online patent-search website that then resulted in perplexing results, perhaps experts here might improve your patent-search skills and/or claims-interpretation skills.– Andreas ZUERCHERCommented Feb 7, 2022 at 17:05
-
Training a neural network on a GPU (using tensorflow or any other package) is a too general concept (or method) and has no novelty as far as I know... I dont think such a general model is patented.– Raul AlvarezCommented May 27, 2022 at 16:57
Add a comment
|
1 Answer
The specific patent EP3129870B1 claims a specific architecture for processing using GPUs. It is limited to what is covered by the claims and by no means covers all applications of machine learning using GPUs. This is far from my field so I can't comment on whether this, in any way, applies to how Tensorflow uses nvidia GPUs. What I can say is that people use Tensorflow GPU acceleration all the time. Amazon and Microsoft provide GPU acceleration of Tensorflow analyses as a service.
This all gets rather complicated. I'd imagine that nvidia provides some level of licensing as part of selling of their GPU products, some of which are designed specifically for machine learning. You might be better off directing your question to the Tensorflow community.
