The Geometry Awareness Function For Neural Networks

[@darksaga2006]

Thanks for the interesting video (and also for mentioning our paper!🎉). I was wondering what you mean that softmax works only in euclidian space? In this paper to some extent we interpret softmax thought a thermodynamics-inspired lens but your manifold-based idea seems interesting. Although the details of your method are a bit fuzzy to me, my understanding is that by embedding in a curved space you are introducing a "non-linear version of temperature", would you agree with this?

[@richardaragon8471]

You are the author?! You are a rock star to me lol. "by embedding in a curved space you are introducing a "non-linear version of temperature"" I do not introduce it. It is already there. I simply take advantage of it and have mapped out what it is and why it is there. docs.google.com/document/d/1qOU_GAzxpG_d-h13DIcgC01_5QEc8QxBSa9WY6kVrAI/edit?usp=sharing

[@darksaga2006]

@@richardaragon8471 haha yes thank you, I am the second author. Pretty interesting stuff, had fun brushing up on gauge theory reg. the connection of potential with space curvature. Keep up the good work!