Publications (39)
Sharpness-Aware Minimization and Muon: Robustness under the Spectral Norm
Wenzhi Zhong, Edward Milsom, Michael Murray
Sharpness-Aware Minimization (SAM) aims to improve generalization by encouraging insensitivity to small, worst-case parameter perturbations. However, the notion of a "small" pertur…
Low Rank Gradients and Where to Find Them
Rishi Sonthalia, Michael Murray, Guido Montúfar
This paper investigates low-rank structure in the gradients of the training loss for two-layer neural networks while relaxing the usual isotropy assumptions on the training data an…
Nodal lengths of eigenfunctions in the disc
Xiaolong Han, Michael Murray, Chuong Tran
In this paper, we derive the sharp lower and upper bounds of nodal lengths of Laplacian eigenfunctions in the disc. Furthermore, we observe a geometric property of the eigenfunctio…
Activation function design for deep networks: linearity and effective initialisation
Michael Murray, Vinayak Abrol, Jared Tanner
The activation function deployed in a deep neural network has great influence on the performance of the network at initialisation, which in turn has implications for training. In t…
Representation Learning for High-Dimensional Data Collection under Local Differential Privacy
Alex Mansbridge, Gregory Barbour, Davide Piras +4
The collection of individuals' data has become commonplace in many industries. Local differential privacy (LDP) offers a rigorous approach to preserving privacy whereby the individ…
Encoder blind combinatorial compressed sensing
Michael Murray, Jared Tanner
In its most elementary form, compressed sensing studies the design of decoding algorithms to recover a sufficiently sparse vector or code from a lower dimensional linear measuremen…
The RHIC Zero Degree Calorimeter
Clemens Adler, Alexei Denisov, Edmundo Garcia +3
High Energy collisions of nuclei usually lead to the emission of evaporation neutrons from both ``beam'' and ``target'' nuclei. At the RHIC heavy ion collider with 100GeV/u beam en…
FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space
Michael Murray, Daphne Chen, Simran Bagaria +7
Pretrained generative robot policies based on flow matching and diffusion have achieved impressive results across a wide range of manipulation tasks. Yet real-world deployments rou…
Towards an understanding of CNNs: analysing the recovery of activation pathways via Deep Convolutional Sparse Coding
Michael Murray, Jared Tanner
Deep Convolutional Sparse Coding (D-CSC) is a framework reminiscent of deep convolutional neural networks (DCNNs), but by omitting the learning of the dictionaries one can more tra…
Kaon Phase Space Density in Heavy Ion Collisions
Michael Murray
The first measurement of kaon phase space densities are presented as a function of transverse mass, center of mass energy and the number of participants. The kaon phase space densi…
Benchmarking Affordance Generalization with BusyBox
Dean Fortier, Timothy Adamson, Tess Hellebrekers +5
Vision-Language-Action (VLA) models have been attracting the attention of researchers and practitioners thanks to their promise of generalization. Although single-task policies sti…
A note on bundle gerbes and infinite-dimensionality
Michael Murray, Danny Stevenson
Let be a bundle gerbe over a fibre bundle . We show that if is simply-connected and the fibres of are connected and finite-dimensional then the Dixm…
Vision-and-Dialog Navigation
Jesse Thomason, Michael Murray, Maya Cakmak +1
Robots navigating in human environments should use language to ask for assistance and be able to understand human responses. To study this challenge, we introduce Cooperative Visio…
On the existence of bibundles
Michael Murray, David Michael Roberts, Danny Stevenson
We consider the existence of bibundles, in other words locally trivial principal spaces with commuting left and right actions. We show that their existence is closely relat…
Scanning the phases of QCD with BRAHMS
Michael Murray, BRAHMS Collaboration
BRAHMS has the ability to study relativistic heavy ion collisions from the final freeze-out of hadrons all the way back to the initial wave-function of the gold nuclei. This is acc…
Benign overfitting in leaky ReLU networks with moderate input dimension
Kedar Karhadkar, Erin George, Michael Murray +2
The problem of benign overfitting asks whether it is possible for a model to perfectly fit noisy training data and still generalize well. We study benign overfitting in two-layer l…
Limiting fragmentation of chemical potentials in heavy ion collisions
Laura A. Stiles, Michael Murray
Thermal models have been used to successfully describe the hadron yields from heavy ion collisions at a variety of energies. For root(S)<17 GeV this has usually been done using yie…
A Few Words Go a Long Way: Language Guided Robot Policy Synthesis
Daphne Chen, Archit Ritesh Jain, Eric Goossen +4
While vision-language-action models have demonstrated impressive zero-shot manipulation capabilities, they remain fundamentally black box policies that are difficult to interpret,…
Bounds for the smallest eigenvalue of the NTK for arbitrary spherical data of arbitrary dimension
Kedar Karhadkar, Michael Murray, Guido Montúfar
Bounds on the smallest eigenvalue of the neural tangent kernel (NTK) are a key ingredient in the analysis of neural network optimization and memorization. However, existing results…
Guarding Against Malicious Biased Threats (GAMBiT): Experimental Design of Cognitive Sensors and Triggers with Behavioral Impact Analysis
Brandon Beltz, Po-Yu Chen, James Doty +16
This paper introduces GAMBiT (Guarding Against Malicious Biased Threats), a cognitive-informed cyber defense framework that leverages deviations from human rationality as a new def…
Centrality and pseudorapidity dependence of the transverse energy flow in pPb collisions at sqrt(s_NN) = 5.02 TeV
Christopher Bruner, Michael Murray
The almost hermetic coverage of CMS is used to measure the distribution of transverse energy as a function of pseudo-rapidity for pPb collisions at TeV. For…
Quasi-periodic paths and a string 2-group model from the free loop group
Michael Murray, David Michael Roberts, Christoph Wockel
In this paper we address the question of the existence of a model for the string 2-group as a strict Lie-2-group using the free loop group (or more generally for compa…
Implicit Bias and Invariance: How Hopfield Networks Efficiently Learn Graph Orbits
Michael Murray, Tenzin Chan, Kedar Karhadker +1
Many learning problems involve symmetries, and while invariance can be built into neural architectures, it can also emerge implicitly when training on group-structured data. We stu…
Mildly Overparameterized ReLU Networks Have a Favorable Loss Landscape
Kedar Karhadkar, Michael Murray, Hanna Tseran +1
We study the loss landscape of both shallow and deep, mildly overparameterized ReLU neural networks on a generic finite input dataset for the squared error loss. We show both by co…
Non associative magnetic translations from parallel transport in projective Hilbert bundles
Jouko Mickelsson, Michael Murray
The non-associativity of translations in a quantum system with magnetic field background has received renewed interest in association with topologically trivial gerbes over $\mathb…
Forward Energy and Multiplicity in Au-Au reactions at sqrt{S}=130GeV
Michael Murray, BRAHMS Collaboration
For relativistic heavy ion collisions the energy flow in the collision reveals information on the equation of state of matter at high density. The BRAHMS experiment has studied the…
Index Theory, Gerbes, and Hamiltonian Quantization
Alan Carey, Jouko Mickelsson, Michael Murray
We give an Atiyah-Patodi-Singer index theory construction of the bundle of fermionic Fock spaces parametrized by vector potentials in odd space dimensions and prove that this leads…
Is there more than one thermal source?
Michael Murray, BRAHMS Collaboration
BRAHMS has the ability to study relativistic heavy ion collisions over a wide range of pT and rapidity. This allows us to test whether thermal models can be generalized to describe…
Studies of the initial and final states of AuAu collisions with BRAHMS
Michael Murray, BRAHMS Collaboration
When heavy ions collide at ultra-relativistic energy, thousands of particles are emitted and it is reasonable to attempt to use hydrodynamic descriptions, with suitable initial con…
HBT in Relativisitic Heavy Ion Collisions
Michael Murray
A summary of current interferometry data in relativistic heavy ions is presented. At sqrt{s}=17GeV a sudden increase in the pion source volume is observed for central PbPb collisio…
Measurement of Mutual Coulomb Dissociation in GeV Au+Au collisions at RHIC
Mickey Chiu, Alexei Denisov, Edmundo Garcia +4
We report on the first measurement of Mutual Coulomb Dissociation in heavy ion collisions. We employ forward calorimeters to measure neutron multiplicity at beam rapidity in periph…
New opportunities at the photon energy frontier
Jaroslav Adam, Christine Aidala, Aaron Angerami +72
Ultra-peripheral collisions (UPCs) involving heavy ions and protons are the energy frontier for photon-mediated interactions. UPC photons can be used for many purposes, including p…
Studying QCD at low x and high mass with CMS
Michael Murray, CMS Collaboration
The CMS heavy ion program can study quark matter over an unprecedented range of Bjorken x and mass. CMS is equipped with excellent detectors to exploit the new physics probes avail…
Flavor Dynamics
Michael Murray
The purpose of BRAHMS is to survey the dynamics of relativistic heavy ion (as well as pp and d-A) collisions over a very wide range of rapidity and transverse momentum. The sum of…
Characterizing the Spectrum of the NTK via a Power Series Expansion
Michael Murray, Hui Jin, Benjamin Bowman +1
Under mild conditions on the network initialization we derive a power series expansion for the Neural Tangent Kernel (NTK) of arbitrarily deep feedforward networks in the infinite…
Bundle Gerbes Applied to Quantum Field Theory
Alan Carey, Jouko Mickelsson, Michael Murray
This paper reviews recent work on a new geometric object called a bundle gerbe and discusses some new examples arising in quantum field theory. One application is to an Atiyah-Pato…
Training shallow ReLU networks on noisy data using hinge loss: when do we overfit and is it benign?
Erin George, Michael Murray, William Swartworth +1
We study benign overfitting in two-layer ReLU networks trained using gradient descent and hinge loss on noisy data for binary classification. In particular, we consider linearly se…
Heavy Ion Physics Program in CMS Experiment
Olga Kodolova, Michael Murray
We present the capabilities of the CMS experiment to explore the heavy-ion physics program offered by the CERN Large Hadron Collider (LHC). The prime goal of this research is to te…
(Anti)Proton and Pion Source Sizes and Phase Space Densities in Heavy Ion Collisions
Michael Murray
NA44 has measured mid-rapidity deuteron spectra from AA collisions at sqrt{s}=18GeV/A at the CERN SPS. Combining these spectra with published proton, antiproton and antideuteron da…