Built a transformer-model replacing traditional feed-forward lawyers by layers inspired by the Kolmogor-Arnold Respresentation Theorem, which states that any multivariate continuous function:

can be expressed as:

where:

• $\phi_q : \mathbb{R} \to \mathbb{R}$ and $\psi_{p,q} : \mathbb{R} \to \mathbb{R}$ are continuous functions
• $x_1, x_2, \ldots, x_n$ are the input variables
• $n$ is the dimensionality of the input space

Used both rational and taylor polynomials as base functions, and compared results with LSTMs and standard transformers. Results weren't amazing, but was an amazing learning experience.

Built a generative adversarial network to generate ququart states from a maximally entangled ququart; in particular:

Designed a gradient estimation procedure to train the generator and achieve high fidelity with the target state. Won 1st place at MIT iQuHack 2024 with my roommates.

Designed a quantum circuit to solve Max-Cut problems with high accuracy. Created Loss Hamiltonians to enforce the constraints of the problem:

where

I also did this with my roommates. Won 2nd place at MIT iQuHack 2025.