Scaling Laws in Learning Properties of Quantum Systems

Imperial College London

Learning properties of quantum systems is a central task in quantum physics and quantum computing. Under the supervision of Roberto Bondesan, I simulated quantum systems at thermal equilibrium and implemented simple machine-learning algorithms to analyze the scaling of sample complexity.

Causal Discovery from Nonstationary Data

Imperial College London

Under the supervision of Wayne Luk, extended constraint-based causal discovery methods to nonstationary settings by relaxing assumptions about causal consistency. Aimed at better understanding causality in time series with applications spanning finance to earth sciences.

Optimized Staple Placement on Helicon

Singh Lab, Rubenstein Group, Altos Labs

Inspired by unfinished realizations duirng my time at Altos Labs and under the supervision of Ritambhara Singh’s Lab and Brenda Rubentstein’s Group, my honors thesis focused on computationally modeling Helicon therapeutics. These innovative therapeutic agents have shown promise in traversing cellular membranes and engaging with previously considered "undruggable" targets, notably within the realm of protein-protein interactions. Leveraging machine learning techniques, I modeled the conformational space of constrained alpha-helical peptides from added hydrocarbon staples, in order to understand optimal staple placement.

De Novo Binder Generation through RoseTTAFoldDiffusion

Altos Labs

Deploying works of RoseTTAFoldDiffusion and Protein Message Passing Neural Network, in collaboration with Simone Bianco’s Computer-Aided Rejuvenation Lab and Adam Frost’s Lab, I generated de novo binders for high-interest targets for Altos Labs. I created novel techniques for validating these candidate binders, altering AlphaFold to scan through thousands of binders with minimal computational expense. These designs and techniques are considered for downstream patents and publications.

Energy Efficiency Prediction

Professor Amy Handlan

Supervised by Amy Handlan, I implemented Neural Network based on open source data from the New York City Department of Energy to create a control group for the energy efficiency of a sample of buildings if Local Law 97 had not been passed (LL97 places a carbon tax on commercial and residential buildings). Preprocessed the data to become a time-series dataset. Tested the effectiveness of LL97 through DiD.

Overfeat: Object Detection

Professor Ritambhara Singh

Investigated object classification and localization based on the seminal work, Overfeat: Integrated Recognition, Localization and Detection Using Convolutional Networks, by Sermanet et al. The Overfeat model achieves state-of-the-art performance on object detection tasks, demonstrating the effectiveness of CNNs for object detection. Since then, numerous variants of CNN-based object detection models have been proposed, each with its own unique set of strengths and weaknesses.

Semantic Analysis: Media's Relationship To The Presidency From Obama To Trump

Professor Amy Handlan

I endeavored to explain the Trumpian influence on the American language, focusing on a corpus of New York Times articles (an influenced and influential publisher) to map the similarities from the pre-to-post-Trump era. Specifically, this project aims to answer the prompt: How the language of the New York Times has changed from President Barack Obama to President Donald Trump based on sentimental and contextual measures? Scraping New York Times articles from President Obama’s term in 2012-2016, and President Trump’s term in 2016-2020, I will perform my own logit-lasso regression, and cosine similarity analysis as well as employ outsourced semantic models from Hugging Face to derive comparative results.

Adverserial Search Bot: Game Of Thrones

I was tasked to develop a Game of Thrones bot for our Aritificial Intelligence class. The project entails building an AI bot to play a grid game (i.e., a game played on a discrete grid). Building an effective bot for this task will require that you explore different techniques: from adversarial search to machine learning (i.e., function approximation) to reinforcement learning to multiarmed bandits.