CV
General Information
Full Name | Hitesh Sapkota |
Languages | English, Nepali |
Education
- 2017-2023
PhD in Computer Science
Rochester Institute of Technology, Rochester, NY, USA
- Developed Distributionally Robust Optimization (DRO) based Bayesian Multiple Instance Learning technique for Anomaly Detection resulting.
- Developed Deep Reinforcement Learning technique for Partial Sketch based Image Retrieval.
- Developed Evidential Openset Detection techniques considering Imbalance Classes and Few Shot Learning Setting.
- Developed Distributionally Robust Ensemble of Sparse Networks for better Calibration, Debiased, and Openset Detection Performance.
- Developed Generalized Focal Loss Ensemble of Low-Rank Networks for Calibrated Visual Question Answering.
- 2012 - 2015
BE in Electrionics and Communication
Institute of Engineering (IOE) Pulchowk, Lalitpur, Nepal
- Performed vehicle number plate detection as a final year project using neural network.
- Financial support from Ncell for excellent academic performance.
Experience
- 2023 DEC - PRESENT
Applied Scientist II
Amazon, Hardware-Product Integrity, Sunnyvale, CA, USA.
- Devising LLMs that can assist engineers in the hardware manufacturing process by providing solutions to the various challenging questions.
- 2022 May - 2022 Aug
Applied Scientist Intern
Amazon, Hardware-Product Integrity, Sunnyvale, CA, USA.
- Designed and Implemented embedding adaptation using attention-based architecture in gas sensor technologies.
- Performance improvement of 6% over the existing baseline.
- 2021 Jun - 2021 Aug
Applied Scientist Intern
AWS, Support, Seattle, WA, USA.
- Designed ML models to detect AWS service failures (issues) early before impacting customers significantly.
- Performance improvement of 4% over the existing baseline.
Honors and Awards
- 2022
- KDD Travel Award
- 2017
- RIT Ph.D. Merit Scholarship
- 2014
- Ncell Scholarship and Excellence Award
Academic Interests
-
Anomaly Detection.
- DRO based Bayesian Multiple Instance Learning.
- Bayesian Nonparametirc Video Paritition Techniques.
- Multiple Instance Active Learning Techniques.
-
Openset Detection.
- DRO based Evidential Learning under Class Imbalanced Setting.
- DRO based Evidential Learning for Biased/Spurious Correlation Setting.
- Fairness AI.
-
Reinforcement Learning.
- Partial Sketch based Imange Retrieval using Double Exploration.
- Dense Object Detection using Evidential Q-Learning.
-
Network Sparisification.
- Calibrated Sparse DNN using Distributionally Robust Ensemble of Lottery Tickets.
- Generalized Focal Loss Ensemble of Multimodal Sparse Network for Visual Question Answeing.
-
Generative AI.
- Sparsified Generative AI models that are sparse, uncertainty aware, and unbiased.
Other Interests
- Hobbies: Reading, Hiking, Traveling.