About me (CV) (Resume)
Sarvagya is a graduate student at University of Massachusetts in Boston, focusing on machine learning. He did an internship at the Boston Children’s Hospital, where he also did his MS Theis. His specific interessts include reinforcement learning, information theory, model compression, theoretical machine learning.
Specific Projects:
A) Internship at Boston Children’s Hospital, Harvard Medical School (Abstract):
- Participated in a pioneering research project aimed at utilizing deep learning methods to analyze interictal Intracranial EEG (iEEG) for improved surgical planning in drug-resistant epilepsy (DRE) patients.
- Conducted data curation of iEEG signals of patients with DRE, including the extraction of relevant features for analysis.
- Employed unsupervised machine learning techniques to analyze iEEG signals and differentiate between signals originating from resected/non-resected parts of the brain.
- Analyzed signals to distinguish between patients who have had successful and unsuccessful surgery.
- Successfully utilized a pretrained VGG16 network to measure the visual complexity of time-frequency (TF) iEEG images, which was hypothesized as an interictal biomarker of epileptogenic zones.
- Author of the thesis: ”Visual Complexity of the Time-Frequency Image Pinpoints the Epileptogenic Zone: An unsupervised Deep-Learning Tool to Analyze Interictal Intracranial EEG” (in the process of being published)
- Work accepted at New England Science Symposium (NESS).
B) Object Detection and collision avoidence for self driving cars:
- A project I worked on at the Indian Institute of Science in Bangalore, India. Under the supervision of Prof. Chiranjib Bhattacharyya, we collected and curated the data by driving around the IISc campus and trained models on them.
C) Equal superposition using reinforcement learning (link):
- This was a code I wrote during my brief time at the Tata Institute of Fundemental Research in Mumbai. I worked on a small RL model to make any quantum state go to equal superpisition.
D) Model Compression:
- Personal project where I worked on a model compression technique. I compresed a model trained on MNIST dataset by 99% (from around 60,000 parameters to 380) with classification accuracy of 85%. Work still in progress (outputs).
Poster Presentation:
A) Time - Frequency Visual Complexity (Poster)
Other Interests
Other interests include playing musical intruments, jogging around Boston, always interested in trying new cuisines and yoga.
Contact
I can be reached via sarvagya.gupta@gmail.com or sarvagya.gupta001@umb.edu and maybe we can chat about ML stuff.
My other profiles: