Siddharth Goel 
- Currently, working as Applied Scientist at
- Former graduate student at
, MSE in Data Science (Dec, 2021)
Areas of Interest
Data Science, Machine Learning, Deep Learning
Research
Completed Masters thesis in Navigation To Multiple Semantic Targets In Novel Indoor Environments [pdf] [slides]
Keywords - Multi-object Navigation, Visual Navigation, Object-goal Navigation
Advised by - Kostas Daniilidis, Georgios Georgakis, and Bernadette Bucher
-
Optimized semantic map prediction in Learning2Map (L2M) algorithm for indoor navigation Pytorch, Python - Explored the various sub-components (map prediction, goal selection, navigation) of training an autonomous agent to navigate to different objects in an indoor environment.
- Improved IoU and F1 score metrics by changing the loss function with Focal Loss, incorporating an LSTM layer in the network, and fine-tuning the weights for different components of the loss objective.
- Experiments performed using Matterport3D dataset on AI Habitat simulator.
Summer Internship (May - Aug, 2021) @
-
Forward Looking Sampling Python, SQL, Amazon Sagemaker, S3 - As an Applied Scientist Intern on the Supply Chain Optimization Technologies (SCOT) - Fulfillment Optimization (FO) team.
- Designed and evaluated a novel forward-looking sampling strategy to generate high-fidelity samples of the millions of customer orders placed on Amazon.com and its subsidiaries.
- Performed K-means clustering to incorporate demand forecast and produce samples which are representative of future demand, leading to better planning and execution of resources for peak periods (Prime Day, Black Friday, Christmas and New Years etc.).
- The samples are used to run simulations based on which Amazon plans its entire order fulfillment supply chain across the world.
Projects
-
US Traffic Accidents Predictive Analytics, Python, scikit-learn, matplotlib, seaborn, Spark Github Repo - Built the complete data science pipeline by performing extensive exploratory data analysis, data pre-processing, feature engineering, and data modelling on about 3 million records of the US Traffic Accidents dataset.
-
Audio Source Separation Deep Learning, PyTorch, Python Github Repo - Separated MUSDB18 dataset mixture tracks into vocals, drums, bass, and other instruments using LSTM and state-of-the-art deep learning models.
Courses taken at UPenn
| Data Science | Deep Learning | NLP | Robotics |
|---|---|---|---|
| CIS 520: Machine Learning | CIS 522: Deep Learning for Data Science | CIS 530: Computational Linguistics | ESE 650: Learning in Robotics |
| ESE 542: Statistics for Data Science | ESE 546: Principles of Deep Learning | ||
| CIS 545: Big Data Analytics | |||
| STAT 535: Time-series Forecasting Methods |
Teaching Assistant
| Course | Role | Semester |
|---|---|---|
| ESE 545: Data Mining | TA/Grader | Spring 2020 |
| ESE 542: Statistics for Data Science (Coursera) | Course Development Assistant | Summer 2020 |
Fond of …
| Food & Drinks | Activities | Places | Animals |
|---|---|---|---|
| Coffee | Swimming | Beaches | Elephants |
| Beer | Cycling | Mountains | |
| Wine | Reading | ||
| Mexican | Cooking | ||
| Pizza |
Shows I liked
| Name | Genre | Platform |
|---|---|---|
| The Last Dance | Sports (NBA) | Netflix |
| Formula 1: Drive to Survive | Sports (F1) | Netflix |
| The Test: A New Era for Australia’s Team | Sports (Cricket) | Amazon Prime |
| Free Solo | Sports (Rock Climbing) | Disney+ |
| The Dawn Wall | Sports (Rock Climbing) | Netflix |
| The Final Table | Food | Netflix |
| The World’s Most Extraordinary Homes | Lifestyle | Netflix |
| Queen’s Gambit | Fiction | Netflix |