2025 USA-NA-AIO Round 1, Problem 2, Part 6

AI Olympiads 2025 USA-NA-AIO Round 1
1

Part 6 (5 points, non-coding task)

We make the following modifications on the previous part.

  • We consider a special symmetric neural network that out_features = in_features.

  • No bias in all affine transformations.

  • The transformation matrix from the hidden layer to the output layer is binded to be the transpose of the transformation matrix from the input layer to the hidden layer

What is the total number of learnable parameters in this model?

  • Reasoning is not required.
2

\color{green}{\text{### WRITE YOUR SOLUTION HERE ###}}

in_features * hidden_features

\color{red}{\text{""" END OF THIS PART """}}