Publications
João Marcos Correia Marques, Ramya Ramalingam, Zherong Pan, Kris Hauser, “Optimized Coverage Planning for UV Surface Disinfection.” (Accepted to IEEE International Conference on Robotics and Automation 2021) [link]
M. Grueter, K. Duran, R. Ramalingam, R. Libeskind-Hadas, “Reconciliation Reconsidered: In Search of a Most Representative Reconciliation in the Duplication-Transfer-Loss Model,” IEEE/ACM Transactions on Computational Biology and Bioinformatics, September 2019, Special issue for the 17th Asia Pacific Bioinformatics Conference, 2019. [link]
Current Research
AMISTAD Lab (Sept 2020 – Present):
I am working under Professor George Montañez in the AMISTAD Lab , a lab that works on theoretical machine learning problems.
Currently, I am in a three-member team studying generalization error bounds of learning algorithms. We are writing a paper describing our findings on these error bounds using algorithm capacity, a variable which measures the mutual information between the distribution across training data sets and the distribution across hypotheses.
Intelligent Motion Laboratory (May 2020 – Present):
This past summer, I worked as a research intern under Professor Kris Hauser at the University of Illinois at Urbana-Champaign. Our research was in the area of intelligent planning for autonomous robots; it was motivated by the Covid-19 pandemic, when it became apparent that automated, reliable and low-cost disinfection systems would be invaluable in the healthcare sphere. We developed an efficient system for autonomous and complete disinfection of complex indoor environments using UV-light carrying robots. In November 2020, we submitted a research paper detailing this work to the IEEE International Conference on Robotics and Automation 2021. As part of the Distributed Research Experience for Undergraduates program, I wrote about how my week-to-week research progressed over the summer, so check out this page for more details about my work.
I am continuing to work with Prof. Hauser and his lab, exploring more involved questions about UV disinfection. For example, what is the most efficient way to disinfect an environment given priority surfaces (high-touch surfaces, low-touch surfaces, etc.), and how do we sample efficiently in a robot’s configuration space (as opposed to physical space) to prevent path redundancies?
Proofpoint (September 2020 – Present):
I am on a Clinic team at Harvey Mudd for the cybersecurity firm Proofpoint. We are working on the problem of machine learning model interpretability. Proofpoint uses ML models to classify threats in emails, websites and other online mediums. Often, analysts who are not domain experts are tasked with explaining model decisions to clients. We are working on an end-to-end system that provides visual explanations to help users and analysts understand why misclassifications are made by a trained black-box model.
We have implemented and are currently analyzing a system that provides explanations for individual predictions made by trained random forests, based on the CHIRPS model developed by Julian Hatwell, Mohamed Medhat Gaber and R. Muhammad Atif Azad. We would like to ensure the explanations themselves are interpretable, and extend/pivot our work to encompass other models, such as K- nearest neighbors and convolutional neural networks.
Past Research
Computational Biology Laboratory (May 2018 – July 2018):
I worked as a research intern under Professor Ran Libeskind-Hadas in the computational biology laboratory at Harvey Mudd. We did algorithmic research in the area of cophylogeny, the study of pairs of species that influence each other’s evolution. A maximum parsimony reconciliation (MPR) represents a set of events that best explains relationships between evolutionary trees of biological pairs. We investigated the problem of how well a single reconciliation may represent a space of equally optimal reconciliations. We developed efficient algorithms to determine different kinds of “representative” reconciliations from a given set, and I worked with another student to implement and test our developed techniques. Our empirical studies demonstrated that single reconciliation does a poor job of representing its MPR space in general.
Our research paper on the topic was accepted to the 17th Asia Pacific Bioinformatics Conference, and was published to the IEEE/ACM Transactions on Computational Biology and Bioinformatics journal.
Internships
Google (May 2019 – August 2019):
I was an Engineering Practicum Intern on the API Client Tools team at Google Cloud Platform. This team generates client libraries in many different programming languages to help developers who call Google APIs in their applications. The client libraries supported server-side JavaScript (Node.js) but not browser JavaScript.
I worked with a fellow intern to enable a fallback protocol which would support browser remote procedure calls (RPCs), by using protocol buffers over HTTP/1.1. This allowed developers to use existing NodeJS libraries in browsers by simply writing Node.js code and bundling the files into a browser package. We enabled our fallback mechanism to be instantiated automatically in browser, allowing RPCs to go through.
Our work was deployed in August 2019, and is still in use today. All of the work I did was open-source; you can view it here.