Improving Mixed Reality for HoloLens, one keystroke at a time:- object and scene mapping- head tracking

New England Research & Development Center (NERD):- Computer Vision: object detection, 2D and 3D semantic segmentation, image captioning- Deep Learning for Audio: event detection- Natural Language Processing: NER and classification- Reinforcement Learning: Evolutionary Strategies, DQN, Contextual Bandits- Recommenders: SAR, Factorization Machines, RNNs for browsingcutting-edge machine learning @ scale

Machine Learning for Microsoft's flagship Azure cloud platform at New England Research & Development Center (NERD):- Azure Machine Learning Studio V1- Microsoft R Server- HDInsight Spark (pySpark, Hive, HDFS/WASB, Azure Data Lake) with H2O- Data Science Virtual Machine- Deep Learning with MXNet on Azure N-series VMs

Spark (MLlib), HDFS, ensemble methods- machine learning credit risk engine for micro-loans (P2P loans)

SVMs, decision trees (forests, boosting), elastic net, feature creation / selection / ranking, pipelining, categorical data- consumer behavior analytics platform in use by Fortune 500 companies- designed custom machine learning and data mining solutions for business owners

regression trees, weighted logistic regression, cross-validation optimization techniques, sparse data- collaborated with the CTO of the company to deliver a superior Real Time Bidding predictive algorithm

multi-step Monte Carlo methods, counterparty default predictive models- development of Initial Margin predictive model for OTC derivatives- risk-adjusted pricing: Potential Future Exposure, Credit & Debit Value Adjustment- development and support of market risk predictive production models

PCA/SVD, factor models, Monte Carlo, risk prediction, time series analysis, curve fitting- created, implemented & productionalized a company-specific market risk predictive model- performed numerous statistical studies which enhanced bank's existing production predictive risk models- defended the validity of newly-built models for Model Validation and Approval- market data and model simulation diagnostics- IT infrastructure development & enhancement

CSC336, Numerical Methods, Department of Computer Science, University of Toronto- floating point arithmetic, numerical integration, numerical solutions of ODEs- prepared and delivered tutorial lectures, held office hours and marked tests and assignments

C (GSL), Matlab, maximum likelihood estimation, PCA- alternative techniques to high-dimensional optimization for quantum chip state tomography- research resulted in Physical Review A journal publication

Matlab, Unix scripting, system administration- developed a Matlab package for visual and statistical analysis of climate model simulations and satellite data