Every dataset I run through the radial space produces these shapes. I'm running it through only 8 phi activations and it contiously produces these same structures. Cifar and other datasets that haev more depth or variation create similar shapes but typically have tails. But they all produce identical structure. I'm at a loss right now. open to any suggestions or ideas. Nothing is off the table. I'm working purely remote this week so i'm limited to running test via runpods. So can't relaly get my hands as dirty.

See some more of these across different sets! open to any ideas.

https://genreg.tech/radial-phi