Brendan Mallery
Tufts University
I am a 5th year PhD candidate in the mathematics department at Tufts University, working with Professors James Murphy and Shuchin Aeron.
My research focuses on:
- Representation learning for probability distributions with optimal transport;
- Geometric approaches to machine learning and data science;
- Interpretability and safety in AI, through the analysis of latent spaces in deep learning models.
- Modeling and analysis of time-evolving networks;
I have also published work in topological data analysis and geometric group theory.
Here are some pics from my research:
Synthesis and Analysis — Generating and reconstructing data as probability measures using entropy-regularized optimal transport.
Dictionary Learning on LLM Document Embeddings — Visualizing unsupervised clustering of dictionary codes.
Robust Signal Recovery with Entropy-Regularized Barycenters — Point clouds with dropout noise can be recovered by projecting onto set of barycenters.
Time-Evolving Curvature of Networks — Evolving cluster structure is detected using time-evolving Ricci curvature.