Creating innovative privacy-enhancing Ads AI products

Accepted Return Offer to join LinkedIn as a Software Engineer in Machine Learning upon graduationReceived a rating of "Beyond Expectations" (5/5)

Joined LinkedIn Marketing Solutions, Ads AI team and conducted a comprehensive analysis of Multi-Task Learning (MTL) for LinkedIn Audience Network (LAN) Web and App CTR Prediction:(Presentation slides attached below, can provide recording upon request)- Designed, implemented, and tuned several MTL models including Hard Sharing, Soft Sharing, Multi-gate Mixture-of-Experts (MMoE), Customized Gate Control (CGC), and Progressive Layered Extraction (PLE) using Keras, Keras Functional API, and Tensorflow (Python)- Significantly improved CTR AUC through designing and implementing a innovative model architecture- Analyzed and compared the various implemented model architectures in terms of performance gain for each task, trainability/robustness, computational cost, thriving on complexity, etc. - Retrieved from Hadoop clusters, sampled, and wrangled datasets of 2 billion+ samples using PySpark - Reviewed advanced theory and experimental analysis of various MTL models and techniques- Documented a 21-page analysis/RFC doc and presented findings, suggestions, and next steps to the team and company, which exceeded project expectations

Developed a simulation model in Python (Matplotlib, Scipy, Numpy), analyzed probabilistic equations, and reviewed advanced literature to identify equilibria in dynamical systems of bacterial composition vectors, in Professor Itsik Pe'er's Computational Genomics Lab

- Analyzed security market and worked alongside CEO to create a solution in identity management, resulting in a startup- Anonymized, wrangled, analyzed, and visualized large data of 10,000+ users in Python (Pandas, Numpy, Matplotlib)

• Studied and applied advanced ML techniques to build and test a pipeline automating the analysis of planarian videos• Co-created, analyzed, and improved the performance of a video classification neural network model by 20%+• Created a library for prediction making and performance analysis, effectively considering the dataset problems• Troubleshooted and analyzed WormPose and nnClassify software packages built using deep learning• Presented results and methodology to the CS department and biology and CS advisors and wrote 30+ reports• Utilized: Python (Keras, Tensorflow, Sklearn, Numpy, Matplotlib), Matlab, Excel, Git/Github, PowerPoint

• Collaborated with a team of 6 to analyze the impact of gerrymandering on precinct boundaries in New Hampshire’s electoral districts using various fairness measures• Co-authored the final paper and presented to the math department resulting in 290+ YouTube views• Helped create Districtr for New Hampshire, an open-source mapping tool developed by MGGG for drawingdistricting plans and analyzing the simulated results of past elections under the plan• Utilized: Python (Pandas, Geopandas, Matplotlib, Numpy, ggplot), PowerPoint

• Assisted and shadowed the communications chair in designing marketing tools