I do not know very much about the topic, but would it be possible to take for example all of each users data, (power distribution, hr, cadence, intervals made, training load etc.) and train a deep learning model to prescribe training (intervals, zone distribution, training load, etc.) so as to optimize the power over a specific duration on a given day (for example make sure I get close to my 20 minute best power performance by July 5th).
With enough users and data wouldn’t this be a revolutionary training bomb?
I can’t tell you if it will or it won’t work but I do know of a major consideration when dealing with adaptive training programs: The quality of your data will equal the quality of the result. These algorithms need to be driven by a set of maximal efforts, which means testing. Take eFTP for example, If you don’t do a maximal effort that is in line with your current FTP or one that would trigger an increase then your eFTP will drop. If you based your training on eFTP alone and didn’t test often enough the training would be rubbish.
That would be cool but tricky to get right as @Bogdan_Rylski points out it all depends on the quality of the inputs. And there are many external factors that are not currently captured in Intervals.icu (sleep quality etc.) but there is a todo list item to add those.