Final papers written at Stanford

I graduated from Stanford in June 2019. Below are some of my final papers.

Computational Biology

Independent Study with Bejerano Lab

During my senior year, I conducted an independent research project under the guidance of Gill Bejerano and Yatish Turakhia. The project centered around the use of short reads of DNA (~150bp) to detect structural variants (hundreds or thousands of base pairs). Approaches to Structural Variant Detection with Short Next-Generation Sequencing Data.

Computer Science

Artificial Intelligence: Principles and Techniques (CS221)

In the past few years, Artificial Intelligence has shown its mettle in all manner of fields. In particular, revolutionary work has been done in the area of abstract strategy games, with AI agents for Chess, Go, and Shogi reaching unprecedented superhuman levels. Beyond providing amusement, game playing provides a valuable opportunity for testing the abilities of high performant artificial intelligence. In this paper, we tackle the two-player abstract strategy game of Hex with the hope of creating a medium level player using enhancements to traditional game playing algorithms such as Minimax and Monte Carlo Tree Search (MCTS). Our findings show, in support of the recent work by DeepMind, that an increasingly sophisticated MCTS is a promising gameplaying algorithm and with enough complexity, has the potential to achieve superhuman strength in a wide variety of games. An AI Agent for Playing Hex.

Convolutional Neural Networks for Visual Recognition (CS231N)

Deep convolutional neural networks are applied to the problem of abnormality detection in musculoskeletal X-rays of the lower extremities. Methods build on prior medical image classification approaches by proposing more sophisticated solutions to the problem of Multiple Instance Learning. Two approaches in particular, one based on recurrent neural networks and the other on Attention models, are shown to outperform static pooling methods. General performance approaches radiologist-level abnormality detection on certain body parts. Recurrent and Attention-Based Approaches to Multiple Instance Learning for Musculoskeletal Abnormality Detection.

Incentives in Computer Science (CS269I)

In this paper we pose the problem of congestion caused by ride-sharing companies as a negative externality. We introduce and analyze previous attempts to curb the externality of congestion in urban areas. We determine that existing methods improve social welfare, but do not go far enough. We discuss three different mechanisms that could be used to price access to roads, and how these mechanisms would mitigate the issue of congestion. We discuss various simulations constructed for the purpose of this paper, and how these simulations ratify the suitability of the mechanisms. We conclude that an iterative tolling method, would most accurately price usage of the road, and be the most directly applicable in the real world. Finally, we discuss various implementation strategies and difficulties that might arise with the implementation of these tolls. Ticket to Ride: An Auction-based Approach to Real-Time Traffic Management.

Decision Making Under Uncertainty (CS238)

In this paper we explore reinforcement learning approaches for packet routing in mesh networks and situations where nodes do not have access to global network information. We also implement extensions to Q-routing in an effort to improve performance, including packet dropping penalties and dual-reinforcement learning. We also propose a hybrid algorithm for transitioning standard protocols to reinforcement learning, mitigating the poor performance of the “learning period” of Q-routing. We find that reinforcement-learning techniques successfully and consistently outperform standard routing protocols for both static and dynamic mesh networks. Applying Reinforcement Learning to Packet Routing in Mesh Networks.

Writing

Fiction Writing (ENGLISH 90)

I’m an avid reader and occasional writer. ENGLISH 90 was the introduction to fiction writing class offered at Stanford, which I took with equal parts trepidation and excitement. I was advised by the brilliant Austin Smith, who guided me through the process of writing my first long short story. The Poison Tree.

Religious Studies

Perspectives on the Good Life (RELIGST 12N)

I took two very consequential classes in my freshman year at Stanford. The first was CS106B, the introduction to computer science class. The second was RELIGST 12N, “Perspectives on the Good Life”. This was an introductory seminar taught by professor Lee Yearley that explored texts concerning purpose, adversity and beauty. The class challenged value judgements that I had formed prior to Stanford and alerted me to novel modes of thinking outside of my Church of England upbringing. Truth.

Chuang Tzu (RELIGST 212)

In RELIGST 12N we gave particular focus to the work of a 4th century BC Chinese philosopher called Chuang Tzu. Lee Yearley offered an entire course of Chuang Tzu, which I took in my sophomore year. We explored fiction by authors such as Herman Melville and Albert Camus, which held interesting parallels to the ancient work. The Talent of a Sage and The Way of a Sage.