Connall Garrod

Connall Garrod is a Research Fellow at the Mathematical Institute, University of Oxford, within the Department of MATH. His research focuses on the application of Random Matrix Theory to Machine Learning, with specific interests in Neural Collapse and Neural Network Hessian Spectra. Garrod is a DPhil Student supervised by Jon Keating, and his recent work includes the paper 'Unifying Low Dimensional Observations in Deep Learning through the Deep Linear Unconstrained Feature Model.' He has received several scholarships, including the Charles Coulson Scholarship (2023-present), the Good Ventures Foundation Scholarship (2022-23), and the Chubb Foundation Scholarship (2016-19).

Connall Garrod is a PhD student at the University of Oxford, specializing in Deep Learning, Machine Learning, and Neural Collapse. His research focuses on understanding the theoretical foundations of deep learning, particularly the phenomenon of neural collapse and its implications for model generalization. Garrod's work explores how deep linear models can unify low-dimensional observations and how cross-entropy dynamics influence learning. His publications analyze the persistence of neural collapse despite low-rank biases and propose methods like Hadamard initialization to improve tractability. His research contributes to the broader understanding of implicit biases in deep learning and their impact on model behavior.