Ashwin Murthy Email & Phone Number

San Diego, Ca, Us

AI Hub Teamhttps://aihub.qualcomm.com/

Seattle, Wa, Us

Acquired by Qualcomm

Mountain View, California, Us

- Wrote a PyTorch implementation of (arXiv:1711.04340), which trains a GAN to help with image classification when data is scarce- GAN takes in an image and outputs a new altered image from the same class.- Passing training data through GAN creates increased sample variation and an improved deep learning classifier- Showed stat-sig improvement on omniglot dataset with <5 samples per class- More details at github link below

- Helped develop a library which uses Bayesian Optimization for hyperparameter tuning of ML models- Used same BO methods to optimize large-scale production systems (e.g. News Feed Recommender system) by tuning hardcoded parameters via A/B tests- Helped build self-serve internal tool enabling teams to optimize their systems with our library- Contributed to open sourcing our library Ax (https://ax.dev/)

- Contributed to end-to-end machine learning stack (pipeline, training, feature computation, inference) for scoring and filtering out low-quality Push, Email, SMS notifications- Helped maintain + add features to C++ backend system for serving the models at 1M+ QPS- Developed models + heuristics to filter out low quality SMS; drastically reduce volume to appease carrier demands while preserving value to users- Experimented with new modeling approaches to predict incremental likelihood of visitation from sending a given SMS- Built out pipelines + logging of notification settings changes to use as counter metrics against making notifications more spammy- Built infrastructure to diagnose which users are reachable via notifications (e.g. valid email, settings on, etc.) and flag the reason if user was not reachable

Ithaca, Ny, Us

- Was on course staff for CS 2110, 2800, 4820 (Data Structures + OO programming, Discrete Math, Algorithms respectively)- Hold Office Hours to help students with assignments as well as prepare for exams- Grade assignments and exams

- Fabricated solar cells using PbS nanocrystals (method described here https://pubs.acs.org/doi/abs/10.1021/nn404457c)- Ran experiments varying conditions of fabrication process to test effect on solar cell efficiency

- Worked on an ML system used to predict closeness between two friends- Integrated the system into Instagram to enable predictions for IG users- Ran some tests using these scores to rank followers lists on profiles

- Developed and published an app on the play store called Super Tic-Tac-Toe, a more complex extension of the traditional game- Implemented a one player AI to play against using a minimax algorithm- Can be downloaded from the play store with the below link

Bachelor'S Degree, Engineering Physics + Computer Science

Education record