Dear all,
I am discussing a collaboration on a deep learning project and I would like your opinion on what licenses are commonly used in machine learning and deep learning in a biomedical setting.
The project (public funding) involves training and validating the algorithm using "public" data (data obtained through state-funded health-care) in order to improve the diagnostics of certain diseases.
We will probably not be confined to research settings for long, as the algorithm might offer a significant improvement of the diagnostics. So its implementation in clinic could be feasible. Also, it would most likely be transferable to other labs in the world.
The algorithm itself is not patentable but can be licensed. What "device"/software the algorithm is embedded into can however be patented.
In view of how the data was acquired, and what the algorithm can possibly offer to the diagnostic power of labs, what license would you recommend?
Should we make a distinction between commercial and non-commercial use?
Would a restriction to non-commercial use be detrimental to research scientists? How?
I am concerned that the current legislations and control agencies lack expertise on these.
Thank you very much for your input.
Julie,
Before going into licensing agreements you might want to take a step back and consider doing a valuation of the IP (does not matter if it is not patentable) and consider if the product is marketable and worth the effort to invest in all the licensing issues to consider.
In terms of valuation you might want to consider how commercially viable is your product, one thing is academic research and another is a product.
Here are some points to consider:
1) what is the risk of having a false positive or false negative (e.g it is not the same to say that a patient has cancer and turns out not to as opposed to not saying it has cancer and the patient has). Also study the liability that the product might have based on this risk
2) does the product easily couples to existing software and hardware or is there overhead(this substantially diminishes the value to your customer since there is additional costs incurred because of an unfinished solution).
3) While algorithms are not patentable there are other parts of the software which might be worth patenting, depending on how it is constructed and the additional data transformations done as well as output of the data.
Should you wish to discuss these or other points feel free to contact me or post the question
Regards
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication