Jimit Majmudar

Jimit

Hi, I'm a PhD candidate in Combinatorics and Optimization at University of Waterloo, where I'm advised by Stephen Vavasis.

My broad interests are in continuous optimization and machine learning. More specifically, I'm working on provable detection of (potentially overlapping) clusters in graphs, which is a fundamental unsupervised learning problem, using tools from convex optimization. In the past, I have also worked on stochastic processes in the context of mathematical sociology.

Before coming to UWaterloo, I obtained my master's degree from University of British Columbia, and bachelor's degree from Indian Institute of Technology (IIT) Bombay, India.

Between December 2020 and April 2021, I interned as an Applied Scientist at Amazon (London, UK) in the Anomaly Detection & Insights team within Prime Video.

[CV]

Publications & Working Papers

Provable Overlapping Community Detection in Weighted Graphs
with Stephen Vavasis
In Proceedings of 34th Conference in Neural Information Processing Systems (NeurIPS 2020)
[Conference ver]   [arXiv]    [Matlab code]     [Python code]

Voter Models and External Influence
with Stephen Krone, Bert Baumgaertner, and Rebecca Tyson
In The Journal of Mathematical Sociology (2020)
[Journal ver]   [arXiv]

Rethinking Noisy Correlation Clustering
with Stephen Vavasis
In preparation.
Theoretical analyses of Correlation Clustering (CC) algorithms typically use a popular generative model for graphs, called the Stochastic Block Model (SBM), which makes various simplifying assumptions such as edge independence. In this work, we propose a novel generative model, using node embeddings, to overcome the drawbacks of SBM, and make progress towards providing theoretical recovery guarantees for CC using the proposed model.