Yamini Vibha Ananth

WaveNet is a generative neural network for generating raw audio waveforms. Due to its autoregressive nature, inference is prohibitively slow. This project optimizes WaveNet’s inference performance through implementation of static and dynamic quantization & provides a CLI tool for easy benchmarking.

Supervised by Dr. Parijat Dube (IBM Research)

Code, IEE Report, Slides

Implementation of an information retrieval system for discovering and extracting structured information from natural language on webpages based on user queries. Benchmarks spaCy and SpanBERT for named entity extraction compared to LLM (OpenAI GPT3.5-turbo).

Code

Podcasts suffer from a discoverability problem; 90% of listeners only consume the top 1% of content. We demonstrate the feasibility a content-based recommender system using text similarity, a collaborative filtering based approach using user data, and a hybrid approach incorporating both.

Supervised by Dr. Chris Wiggins (Chief Data Scientist, NYT) for Applied Math Senior Seminar

Code, Slides, Article

This project extracts correlations between tree and temperature data in NYC by implementing the Apriori algorithm for finding association rules over an integrated dataset created from NYC OpenData’s Hyperlocal Temperature Monitoring dataset—temperature data from neighborhoods with the highest heat mortality risk during the summers of 2018 and 2019—and the 2015 Street Tree Census dataset from the NYC Parks & Rec department.

Code

An economic analysis and optimization of type, quantity and positioning of drones to fight Australian bushfires. Developed ODE model, linear optimization model, least-costs path discovery algorithm, and economic price modeling over 10 year period.

Supervised by Dr. Daniel Lacker for the COMAP International Mathematical Competition in Modeling 2021

International Honorable Mention

Paper