Explanation of 1 bit for the invertibility

[Vitaly-Protasov]

When discussing the saving of additional information for making f invertible, you mentioned only one bit per element. It would be helpful to explain what this additional information entails and how it contributes to making f invertible.
Thanks!

[fsherry]

For both GELU and SiLU, you can split the x-axis into a left and right part, so that the functions are invertible when restricted to either part. The extra bit tells you whether the input is in the left or right part. For general activation functions you will need to use more than one bit.