Welcome to my homepage! My name is Himalaya Sharma and I am a Data Scientist at CARFAX Canada, where I lead AI initiatives involving large-scale data decoding to solve business problems. I completed my Master of Engineering in Electrical & Computer Engineering (specializing in AI & Machine Learning) at the University of Waterloo, and my undergraduate degree at Birla Institute of Technology and Science with a double major in Electronics & Communication and Biological Sciences.
Previously, I worked as a Data Scientist at the Ubiquitous Health Technology Lab (UbiLab) at University of Waterloo, and as a Research Assistant in computer vision at the Image and Vision Computing Lab. I have also been a research intern at New York University Abu Dhabi and Vienna University of Technology, where I worked on Deep Learning and time-series modeling under Prof. Muhammad Shafique.
Download my resumé.
- Artificial Intelligence
- Data Science
- Product Sense
- Statistics
Master of Engineering in Electrical & Computer Engineering (Specialization in Artificial Intelligence & Machine Learning), 2022
University of Waterloo
Bachelor of Engineering in Electronics and Communication Engineering, 2016
Birla Institute of Technology and Science, Pilani
Master of Science in Biological Sciences, 2016
Birla Institute of Technology and Science, Pilani
News
Experience
- Led and currently leading AI initiatives involving large-scale data decoding to solve business problems.
- Performed Human Activity Recognition (HAR) for 20+ scenarios, achieving a mean F1 score of 0.81, by utilizing both classical and ensemble multi-class classification models.
- Automated data loading and annotation for 40+ participants, obtaining a 70% time reduction, by leveraging Python scripts to extract sensor data from Web APIs, JSON & Excel files and SQL dumps.
- Constructed a scikit-learn data pipeline to facilitate sequential loading, wrangling, feature engineering, and modeling, resulting in a 2x faster experiment iteration, followed by deployment on Azure DS Virtual Machine.
- Identified 4 deep learning models for performance evaluation, by conducting a literature survey of state-of-the-art Image Quality Assessment (IQA) algorithms.
- Wrote bash scripts to benchmark PyTorch & TensorFlow models on 5+ authentic & synthetic distortion image databases.
- Deployed batch jobs on multi-GPU compute nodes on High Performance Computing (HPC) clusters, resulting in a 50% reduction in overall training and evaluation time.
- Implemented data generators to facilitate real-time data feeding to models, resulting in a 80% reduction in memory consumption.
- Leveraged Google’s attention-based Temporal Fusion Transformer (TFT) to improve forecasting performance by 37%, compared to established baselines using classical methods from Python’s Statsmodels.
- Performed literature survey of parametric and non-parametric (machine learning/deep learning) time-series modeling techniques.
- Built a data formatter to filter, preprocess, and transform data, resulting in a 30% increase in productivity by utilizing object-oriented programming principles.
- Experimented with 10+ statistical and machine learning based time-series models for pre-emptive arrhythmia forecasting using electrocardiogram data.
