The first paper is super-interesting. I've thought this would be possible for a bit - although I have to question why only one subject as the training data will be relatively fixed to one swing rather than across multiple, especially as this won't produce (much) deviance in path and also will have a relatively fixed form of kinematics (smooth versus harsh transitions etc). I'd expect a NN to perform well of one player's data, but I wonder how well a model would generalise...
Hi Jonathan, thanks for your reflection. Your questions are valid. I could imagine they started with one swing simply to use the research as a "proof of concept", and then use that to justify further research. However, this is me just guessing, as I wasn't involved in the research. Keep up your great work on your substack!
The first paper is super-interesting. I've thought this would be possible for a bit - although I have to question why only one subject as the training data will be relatively fixed to one swing rather than across multiple, especially as this won't produce (much) deviance in path and also will have a relatively fixed form of kinematics (smooth versus harsh transitions etc). I'd expect a NN to perform well of one player's data, but I wonder how well a model would generalise...
Hi Jonathan, thanks for your reflection. Your questions are valid. I could imagine they started with one swing simply to use the research as a "proof of concept", and then use that to justify further research. However, this is me just guessing, as I wasn't involved in the research. Keep up your great work on your substack!
Thank you!