arXiv subjects

Alex Gower

Publications and source records attributed to Alex Gower.

4 recordsLinked to original sources

arXiv2026

FlowEqProp: Training Flow Matching Generative Models with Gradient Equilibrium Propagation

We introduce Gradient Equilibrium Propagation (GradEP), a mechanism that extends Equilibrium Propagation (EP) to train energy gradients rather than energy minima, enabling EP to be applied to tasks where the learning objective depends on the velocity field of a convergent dynamical system. Instead of fixing the input during dynamics as in standard EP, GradEP introduces a spring potential that allows all units, including the visible units, to evolve, encoding the learned velocity in the equilibrium displacement. The spring and resulting nudge terms are both purely quadratic, preserving EP's hardware plausibility for neuromorphic implementation. As a first demonstration, we apply GradEP to flow matching for generative modelling - an approach we call FlowEqProp - training a two-hidden-layer MLP (24,896 parameters) on the Optical Recognition of Handwritten Digits dataset using only local equilibrium measurements and no backpropagation. The model generates recognisable digit samples across all ten classes with stable training dynamics. We further show that the time-independent energy landscape enables extended generation beyond the training horizon, producing sharper samples through additional inference-time computation - a property that maps naturally onto neuromorphic hardware, where longer relaxation yields higher-quality outputs. To our knowledge, this is the first demonstration of EP training a flow-based generative model.

cond-mat.dis-nn↗

arXiv2025

Learning at the Speed of Physics: Equilibrium Propagation on Oscillator Ising Machines

Physical systems that naturally perform energy descent offer a direct route to accelerating machine learning. Oscillator Ising Machines (OIMs) exemplify this idea: their GHz-frequency dynamics mirror both the optimization of energy-based models (EBMs) and gradient descent on loss landscapes, while intrinsic noise corresponds to Langevin dynamics - supporting sampling as well as optimization. Equilibrium Propagation (EP) unifies these processes into descent on a single total energy landscape, enabling local learning rules without global backpropagation. We show that EP on OIMs achieves competitive accuracy ($\sim 97.2 \pm 0.1 \%$ on MNIST, $\sim 88.0 \pm 0.1 \%$ on Fashion-MNIST), while maintaining robustness under realistic hardware constraints such as parameter quantization and phase noise. These results establish OIMs as a fast, energy-efficient substrate for neuromorphic learning, and suggest that EBMs - often bottlenecked by conventional processors - may find practical realization on physical hardware whose dynamics directly perform their optimization.

cs.LG↗

arXiv2025

How to Train an Oscillator Ising Machine using Equilibrium Propagation

We show that Oscillator Ising Machines (OIMs) are prime candidates for use as neuromorphic machine learning processors with Equilibrium Propagation (EP) based on-chip learning. The inherent energy gradient descent dynamics of OIMs, combined with their standard CMOS implementation using existing fabrication processes, provide a natural substrate for EP learning. Our simulations confirm that OIMs satisfy the gradient-descending update property necessary for a scalable Equilibrium Propagation implementation and achieve $\sim 97.2\pm0.1\%$ test accuracy on MNIST and $\sim 88.0\pm0.1\%$ on Fashion-MNIST without requiring any significant hardware modifications. Importantly, OIMs maintain robust performance under realistic hardware constraints, including 10-bit parameter quantization, 4-bit phase measurement precision, and moderate phase noise that can potentially be beneficial with parameter optimization. These results establish OIMs as a promising platform for fast and energy-efficient neuromorphic computing, potentially enabling energy-based learning algorithms that have been previously constrained by computational limitations.

cond-mat.dis-nn↗

arXiv2024

Saddles-to-minima topological crossover and glassiness in the Rubik's Cube

Slow relaxation and glassiness have been the focus of extensive research attention, along with popular and technological interest, for many decades. While much understanding has been attained through mean-field and mode-coupling models, energy landscape paradigms, and real-space descriptions of dynamical heterogeneities and facilitation, a complete framework about the origins and existence of a glass transition is yet to be achieved. In this work we propose a discrete model glass-former, inspired by the famous Rubik's Cube, where these questions can be answered with surprising depth. By introducing a swap-move Monte Carlo algorithm, we are able to access thermal equilibrium states above and below the temperature of dynamical arrest. Exploiting the discreteness of the model, we probe directly the energy-resolved connectivity structure of the model, and we uncover a saddles-to-minima topological crossover underpinning the slowing down and eventual arrest of the dynamics. We demonstrate that the smooth behaviour in finite-size systems tends asymptotically to a sharp step-change in the thermodynamic limit, identifying a well-defined energy threshold where the onset of stretched exponential behaviour and the dynamical arrest come to coincide. Our results allow us to shed light on the origin of the glassy behaviour, and how this is resolved by changing connectivity upon introducing swap-moves.

cond-mat.dis-nn↗