MSB 318: Mathematical Sciences Building, Room 318

Our [Colloquium](https://sciences.ucf.edu/math/colloquium/) series offers a diverse platform for research scholars, faculty, students, and industry experts to share and exchange ideas, fostering discussion and networking across mathematics, statistics, and data science.

Our own Professor [Christian Kummerle](https://sciences.ucf.edu/math/person/christian-kummerle/) will speak at this week's colloquium on "Iterative Quadratic Reweighting for Low-Rank Neural Network Training & Minimum-Cost Flow Problems."

Abstract: Parameter-efficient training, based on low-rank optimization, has become a highly successful tool for fine-tuning large deep learning models. However, these methods fail at low-rank pretraining tasks, where maintaining both the low-rank structure and the training loss remains challenging.

In this talk, we present recent work on the Quadratic Reweighted Rank Regularizer (Q3R), which leads to a novel low-rank inducing training strategy inspired by the iteratively reweighted least squares (IRLS) framework. The strategy includes an adaptation of the state-of-the-art AdamW optimizer [Loshchilov, Hutter 2018] to Q3R. Unlike other low-rank training techniques, Q3R is able to train weight matrices with prescribed, low target ranks of models that achieve comparable predictive performance as dense models, with small computational overhead, while remaining fully compatible with existing architectures.

In the second part of the talk, we present ongoing work on a novel continuous solver for non-convex minimum cost network problems such as the fixed-charge network flow problem. Such problems arise in multipart pathfinding problems in the Lightning Network, the leading layer-two scaling technology for the Bitcoin protocol.

This talk is based on joint work with Ipsita Ghosh, Sindura Saraswathi, Ethan Nguyen and Andrew Chen.

Speaker Bio: Dr. Christian Kummerle joined UCF's School of Data, Mathematical and Statistical Sciences (SDMSS) as an assistant professor in 2025 and is affiliated with the UCF Institute of Artificial Intelligence. His research interests are in the mathematical foundations of machine learning and the development and analysis of efficient algorithms for large-scale data analysis. His research leverages continuous optimization to address computational and statistical challenges arising from data models involving graph, sparsity and low-rank structures, leading to scalable algorithms with provable guarantees. Dr. Kummerle was a postdoctoral fellow at Johns Hopkins University from 2020 to 2022, an assistant professor of computer science at the University of North Carolina at Charlotte between 2022 and 2025, and received a doctoral degree in mathematics from Technical University of Munich. His research has been published at premier venues in machine learning (JMLR, ICML, NeurIPS) and in mathematics.