Kenyi Takagui-Perez

Effect of interatomic repulsion on Majorana zero modes in a coupled quantum-dot superconducting-nanowire hybrid system

Work with Armando Aligia. We study the low-energy eigenstates of a topological superconductor wire modeled by a Kitaev chain, which is connected at one of its ends to a quantum dot through nearest-neighbor (NN) hopping and NN Coulomb repulsion. Using an unrestricted Hartree-Fock approximation to decouple the Coulomb term, we obtain that the quality of the Majorana end states is seriously affected by this term only when the dependence of the low-lying energies with the energy of the quantum dot shows a “diamond” shape, characteristic of short wires. We discuss limitations of the simplest effective models to describe the physics. We expect the same behavior in more realistic models for topological superconducting wires.

(2024) Physical Review D [Journal Link] [arXiv]

A note on an inversion algorithm for vertical ionograms for the prediction of plasma frequency profiles

Work with Marco Milla. Building upon the concept of utilizing quasi-parabolic approximations to determine plasma fre- quency profiles from ionograms, we present a refined multi-quasi-parabolic method for modeling the E and F layers. While a recent study [AIP Advances 14, 065034 (2024)] introduced an approach in this direction, we identified several inaccuracies in its mathematical treatment and numerical results. By addressing these issues, we offer a clearer exposition and a more robust algorithm. Our method assumes a parabolic profile for the E layer and approximates the F layer with a series of concatenated quasi-parabolic segments, ensuring continuity and smoothness by matching deriva- tives at the junctions

(Fall 2024) In Submission [arXiv]

Holographic Entanglement Entropy

My undergraduate thesis in Physics under my advisor, Profesor Pablo Bueno. Provides an overview of entanglement entropy in quantum mechanics, quantum field theory and the Ryu-Takayanagui Conjecture.

(2022) Pontificia Universidad Catolica del Peru · Senior Thesis [PDF]

Arithmetic coding with GPT to solve the commaAI compression challenge

in progress

Implementation of ICML 2022 sub. Language Driven Semantic Segmentation

My implementation of the "Lseg: language driven semantic segmentation" paper by Boyi Li et al. A dense prediction transformer (DPT) with a modified head encodes at pixel level and a CLIP model that encodes words get combined in a multimodal latent space provides flexibility in image annotation.

[Repository]

Monte Carlo Tree Search for Connect4

I wanted to learn Reinforcement Learning so I thought this would be a good first exercise. The environment of Connect4 consist of a grid in which players take turns to drop disks and the first one to connect 4 wins. With a MCTS depending on how deep we allow the tree to go the algorithm can beat us with clever moves or loss with very dumb ones.

[Repository]

Reproducing results from the Neural Style Transfer paper by Gatis et al.

We have an style image whose texture we want to extract and subsequently embed it into a target image. Using CNNs (usually a Vgg-16/19) this algorithm allows us to produce new images that combine the content of an arbitrary photograph with the appearance of numerous well known artworks. Gram matrices are used to get agnostic spatial information about the images.

[Repository]

ionogramNET: ionospheric echo detection with convolutional neural networks

Replicated the results described from the (closed source) paper "Ionospheric Echo Detection in Digital Ionograms Using Convolutional Neural Networks" from the Advancing Earth and Space Science Journal submissions 2021. The method is based on convolutional neural networks to extract ionospheric echoes from digital ionograms. From the extracted traces, ionospheric parameters can be determined and electron density profile can be derived.

[Repository]