Gauri Gupta, Ritvik Kapila, Ayush Chopra, Ramesh Raskar. First 100 days of pandemic; an interplay of pharmaceutical, behavioral and digital interventions -- A study using agent based modeling [paper] [poster] (AAMAS 2024)
Gauri Gupta, Ritvik Kapila, Keshav Gupta, and Ramesh Raskar. "Domain Generalization in Robust Invariant Representation", presented at International Conference on Learning Representations workshop [paper] [code] [slides] (ICLR 2023)
Nojan Sheybani, Zahra Ghodsi, Ritvik Kapila, and Farinaz Koushanfar. "Zero Knowledge Right of Ownership for Neural Networks (ZKROWNN)", To appear at Design Automation Conference [paper] (DAC 2023)
Recent Research Directions:
Research Experience at UC San Diego:
I am currently working as a Graduate Student Researcher on deep learning and time-series zero-shot learning with the Microelectronic Embedded Systems Laboratory, guided by Prof. Rajesh Gupta at UCSD. At UCSD, I have also previously worked on privacy-preserving ML with the Adaptive Computing and Embedded Systems (ACES) research group, guided by Prof. Farinaz Koushanfar at UCSD. My research involved designing a non-interactive zero-knowledge proof system for neural network watermarking. The research has been accepted at the Design and Automation Conference (DAC 2023) and we've also filed a patent for the same.
Research Experience in collaboration with Camera Culture group, MIT Media Lab:
I worked on “Domain Generalization in Robust Invariant Representation” in collaboration with Camera Culture Group, MIT Media Lab in Winter 2023. The project explored the generalization of invariant representations to out-of-distribution data, investigating whether models invariant to transformations in one domain remain invariant in unseen domains. Our work was accepted in the PML4DC workshop at the International Conference on Learning Representations (ICLR) 2023.
Projects:
In collaboration with Camera Culture Group, MIT Media Lab
Duration: Jan'23 - Mar'23
Investigated generalization of invariant transformations on out-of-distribution data for object detection. Introduced transfer learning for generating data-agnostic representations invariant to group transformations. Demonstrated VAE that learns robust unstructured latent features; Crucial in resource-constrained settings.
Privacy Preserving Machine Learning
Guide: Prof. Farinaz Koushanfar, UC San Diego (Jun'22 - Dec'22)
Developed a Non-interactive Zero-Knowledge system for Neural Network watermarking in a legal setting. Achieved sub-second proof time where WM is extracted from network activation maps using specific triggers. Evaluated ZKROWNN on Multi-layered Perceptron for MNIST dataset and optimized runtime to 29.4 ms.
Navigo: Autonomous Navigation System
Robotics Research project [consolidated report]
Guide: Prof. Henrik I. Christensen, UC San Diego (Sep'22 - Dec'22)
Deployed an end-to-end autonomous driving pipeline with Perception, Planning, and Control stages in ROS. Implemented SLAM for robot localization and environment mapping; developed PID controller for movement. Built the path planning node using A-star search algorithm with maximum safety and min distance heuristics.
Channel Estimation using Deep Learning
Bachelors Thesis [report]
Guide: Prof. Manav Bhatnagar, Indian Institute of Technology Delhi (Jul'20 - Dec'20)
Estimated the channel matrix using pilot signals and achieved higher performance than the conventional methods including Least squares and Minimum mean squared error on the grounds of accuracy and time complexity.
Tree-Structured Neuron Classification
Formulated a constructive algorithm for binary nonlinear classification in multi dimensional input space. Designed linear programming framework for optimization & used twin SVM classifier to tackle class imbalance.
Network Automation using Machine Learning
Research Fellowship [code]
Award by R&D Unit, IIT Delhi and Nokia Communication Technologies (Dec'19 - May'20)
Developed automated configurable XML database parser for live integration with Nokia framework. In a parallel project, designed an LSTM based optical character recognition framework to perform graph digitization in real time.
Big Data Analysis for Europa Smart Buildings
Research Internship [code]
Guide: Prof. Frédéric Le Mouël, CITI Lab Research Group, France (Sep'19 - Dec'19)
Implemented pattern recognition and modelled energy usage time series sensor data in smart buildings. Formulated spatio-temporal summarization and obtained 94.8% accurate dependency on room location, date and time.
Reconfigurable Beacon
Embedded Systems Design Project [blog]
Guide: Prof. M. Balakrishnan, Demonstrated at Open House'19, IIT Delhi (Jan'19 - May'19)
Configured bluetooth low energy modules as economical, compact beacons for use in the AIIMS way-finding project. Developed android app to configure the beacon using AT commands for modification of beacon services.