How newbie and startups make sure they are not violating existing deep learning patents ? what goes in deep learning patents

To explain my query let me take an example ,
one of the know problem in medical imaging is identifying Diabetic Retinopathy and it is being solved by using deep learning methods. And some patents also are on this.
if some one wants to develop a application for this problem using their own deep learning network architecture or using an opensource licensed pretrained ones how to make sure they are not violating existing patents.
in summary, what is that one thing that should be unique so that it wont be treated as patent violation especially when solving problems using deep learning in the filed of medical imaging.

Kindly share any info if you are familiar with this topic.
Thank you

You should consult an IP lawyer about this (I’m not one).

A common position among developers is that you should just go ahead and build the thing you’re interested in and not look at any patents at all. Because if you do look at patents and you’re at some point found to be infringing, it count as willful infringement, for which the other party can claim more damages than for accidental infringement.

Not sure how good this advice is, but it’s one way to approach this issue (i.e. ask for forgiveness, not permission).

If you do decide to look at patents, the only thing that counts is the claims section. If a patent’s claims depends on doing A, B, and C but you’re only doing A and B, then you’re not infringing the patent.

But again, do the smart thing and get legal counsel.

1 Like

Thank you very much @machinethink for explaining it so clearly.