Conducted machine learning research under Dr. Subbarao Kambhampati in the Yochan Lab.Although one expects Generative Adversarial Networks to replicate the distribution of the data they are trained on, in real-world settings with limited data and finite network capacity, GANs suffer from mode collapse. I studied how the non-uniformity of the training data affects the generated distribution, as data from online social media platforms or the web are almost never balanced. In settings where data exhibits bias along sensitive latent axes (eg. gender, race), failure modes of Generative Adversarial Networks (GANs) exacerbate the biases in the generated data.

As a graduate TA for CSE 471: Introduction to Artificial Intelligence, I held two hours of office hours each week to assist students with written and programming assignments that apply theoretical concepts: search, games, Markov Decision Processes, reinforcement learning, logic, neural networks, Bayesian learning, and classical planning. I also helped set midterm exam questions and graded student coursework.

Conducted machine learning research under Dr. Subbarao Kambhampati in the Yochan Lab.Deep learning for computer vision applications requires massive amounts of data. Generating synthetic data via GANs has proven tremendously popular to alleviate this problem. We showcase the inherent flaws in this technique: GANs can perpetuate and further biases in the data. We trained a GAN to generate faces of engineering researchers; through human-subject study, we found the number of faces with feminine features and non-white skin tones decreased in the synthetic data.

As a UGTA for Data Structures and Algorithms (CSE 310) in Fall '17, I led recitations for review of subject material from lectures and walked through example problems to strengthen course concepts. This included, notably, constructing an algorithm to count the number of inversions in an array of numbers iteratively and recursively (using a Divide and Conquer method), and teaching how to approach Dynamic Programming problems.For Principles of Programming Languages (CSE 340) in Spring '18, I held five hours of office hours each week to assist students with programming assignments that apply theoretical concepts. This included, notably, coding a non-deterministic finite automaton to generate a regular language, storing a grammar and calculating first and follow sets, checking syntax and semantics errors by implementing a recursive-descent parser and efficiently processing variables of a program, and building a compiler for a programming language specified by a context-free grammar.

Perfecting advertisers' experience as part of the Talent Media team in LinkedIn Marketing Solutions.Advertisers bid on ad space in a virtual auction. LinkedIn has not been recommending effective bids to Dynamic Ads advertisers who choose to pay by the click; we calculate their minimum bid with a global hard-coded value for click-through rate instead of customizing it to the campaign's objectives or targeting criteria.I used a dynamic forecasted value for this click-through rate to provide optimal bid suggestions tailored to the ad campaign, ensuring they are high enough for campaigns to win auctions, but not so high that the advertisers spend more than they need.

One of two Microsoft Student Partners for Arizona State UniversityThough tech as a field is now growing in breadth and popularity, the areas of interest within it appear to be esoteric and elusive as ever. As an MSP, my primary role was to teach workshops to engage students of all backgrounds to increase their exposure to and develop their skills in up-and-coming fields. My most memorable workshop was Introduction to Machine Learning with TensorFlow.I was invited as one of ten US Microsoft Student Partners to attend Microsoft Build and the global MSP Summit in 2018.

Tutoring is effective because tutors are relatable. I am skilled in guiding students on course concepts, focusing on foundational computer science and mathematics. In addition, I planned and led exam- and concept-review sessions (Calculus for Engineers, Discrete Mathematical Structures).

Intern for the Dynamic Ads team on LinkedIn Marketing Solutions, a two-billion-dollar ads business.Businesses previously saw a fake preview when creating an ad campaign -- this means businesses would not see exactly what they were paying for and that there was a duplication of effort and risk of inconsistency when ad designs changed. This was the top user complaint for Campaign Manager and 60% of users who started creating a campaign never finished purchasing.To ameliorate this problem, I implemented live preview of dynamic ads on the existing Rendering multiproduct using Java 8 as a Functional Programming language, and I wrote a testplan for/effected the testing of the new service. Dynamic Ads with the live preview are available now, and are a key component of Campaign Manager's Objective-Based Buying.

Intern for the Enterprise Money Movement Team in Enterprise Infrastructure Systems. Person-to-person bank transfers were revolutionized with the introduction of Zelle, allowing users to transfer funds immediately to one another, across a network of banks in the United States, including those not their own. USAA launched Zelle Send Money on its mobile app a few weeks into my internship in June 2017, but had displayed a "Coming Soon" label for Zelle Request Money. Using Agile methodologies, I designed, developed, and tested a fully-functioning Java REST API for Request Money. Notably, I wrote REST-to-SOAP calls and implemented concurrent processes. I tested using JUnit and Mockito on a Linux-hosted runtime environment. USAA's Zelle Request Money is currently live through its mobile banking app.