I am a senior pursuing my B.S. in Computer Science at Stanford University. I focus on computer vision, specifically in machine learning and infrastructure to make optimized large scale processing of visual data. I am currently doing research under Prof. Pat Hanrahan in Graphics Lab at Stanford where I am working on developing an optimized astronomy pipeline for processing datasets with large images. I have previously done research under Prof. Sigrid Close in Space Environment and Satellite Systems (SESS) lab in the Aero/Astro department at Stanford University where I worked on optimizing pipeline for meteor detection in grayscale video data. Reinforcement learning also interests me and I see it as an important tool in the advancement of artificial intelligence.

In my free time, I love to read papers and articles on astronomy. I have previously worked with several observatories on detection of asteroids, supernovas, variable stars and comets. I discovered a comet, initially named SOHO 2333 and later recognized by Minor Planet Center as C/2012 N1, using the data from SOHO observatory (more). I am an amateur photographer interested in nature, specifically landscapes and wildlife. Some of my photographs are available here. I love playing chess and reading chess books analyzing classical games.


Automated Image Timestamp Inference Using Convolutional Neural Networks

In this paper, we propose methods to classify the time taken of a picture in two ways: using user submitted tags, namely “morning”, “afternoon”, “evening” and “night” and four time buckets (i.e. 12 AM to 6 AM, 6 AM to 12 PM, etc.). Among the prediction models used were vanilla SVMs and their variants, along with Convolutional Neural Networks ranging from three layer architectures to deeper networks, namely, AlexNet and VGGNet.

Paper Poster

Identifying Injury-Inducing Factors in Baseball Pitchers

This paper aims to identify which metrics in baseball are meaningful in predicting whether a pitcher will get injured before a particular season. We use feature filtering techniques such as PCA and mutual information analysis todetermine which features provide the most useful information regarding whether a player will get injured in the context of a season. We define a cost function and build a framework that takes into account the asymmetric reward and cost associated with correctly (or incorrectly) predicting injuries. We then vary the sets of features on various machine learning techniques (i.e., SVM and Linear Regression) to test each model within our framework and compute a utility score for each model.

Paper Poster

Improved Genetic Algorithms For Reinforcement Learning

In this paper, we apply a variation of genetic algorithms, called grammatical evolution, to a single-agent and a multi-agent benchmark problem. Rock- Sample problem is used as the single-agent problem, and baseline scores are generated from common POMDP solvers as well as two heuristic expert-provided policies. For the multi-agent problem, a novel problem is proposed, called BorderCross, which models a multi-robot coordination problem.