aiqbench analysis
This research advances physical neural networks by embedding trainable nonlinear functions directly into analogue connection hardware, enabling efficient continuous control for tasks like robotics and power tracking. While the approach offers significant node reduction for smooth-valued targets compared to standard multilayer perceptrons, it does not provide parameter efficiency for classification problems with sharp decision boundaries. The study demonstrates that these benefits stem from architectural design rather than specific device types, as validated across CMOS and memristive simulations.
Low-power analogue neural networks with trainable nonlinear connections for continuous control
arXiv:2606.23742v2 Announce Type: replace Abstract: Physical neural networks promise low-power machine learning by computing directly with analogue device physics, but most architectures force nonlinear device responses to act as scalar weights. Inspired by Kolmogorov-Arnold networks, we place trainable nonlinear functions on the connections, making each physical connection a learnable computational element. Realising these functions as analogue band-pass filters on field-programmable analogue arrays, we find that the benefit is task-dependent and follows from the smoothness of the physical basis: the networks represent smooth, continuously valued targets, including robotic kinematics, continuous control, and photovoltaic maximum-power-point tracking, with far fewer nodes and connections than multilayer perceptrons, but offer no parameter-efficiency advantage on classification-like decision boundaries. Trained networks transfer to hardware across approximately 35,000 connections with quantified fidelity, and a dedicated CMOS implementation is projected to operate at approximately 30 microwatts. A memristive realisation reproduces the same behaviour in simulation, indicating that the advantage comes from placing trainable nonlinearity on connections, rather than from a particular device.
Authors:Ian T. Vidamour, Fernando Aguirre, Thomas J. Hayward, Matthew O. A. Ellis, Charles Swindells, Alexander McDonnell, Martin Trefzer, Finley Robins, Luca Manneschi, Susan Stepney, Tony Kenyon, Oliver J. Sutton, Jack C. Gartside, Ivan Y. Tyukin, Adnan Mehonic, Eleni Vasilaki
Abstract:Physical neural networks promise low-power machine learning by computing directly with analogue device physics, but most architectures force nonlinear device responses to act as scalar weights. Inspired by Kolmogorov-Arnold networks, we place trainable nonlinear functions on the connections, making each physical connection a learnable computational element. Realising these functions as analogue band-pass filters on field-programmable analogue arrays, we find that the benefit is task-dependent and follows from the smoothness of the physical basis: the networks represent smooth, continuously valued targets, including robotic kinematics, continuous control, and photovoltaic maximum-power-point tracking, with far fewer nodes and connections than multilayer perceptrons, but offer no parameter-efficiency advantage on classification-like decision boundaries. Trained networks transfer to hardware across approximately 35,000 connections with quantified fidelity, and a dedicated CMOS implementation is projected to operate at approximately 30 microwatts. A memristive realisation reproduces the same behaviour in simulation, indicating that the advantage comes from placing trainable nonlinearity on connections, rather than from a particular device.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Hardware Architecture (cs.AR)
arXiv:2606.23742 [cs.LG]
(or arXiv:2606.23742v2 [cs.LG] for this version)
From: Eleni Vasilaki D.Phil. [view email] [v1] Sun, 21 Jun 2026 15:55:23 UTC (13,966 KB) [v2] Tue, 29 Sep 2026 12:59:42 UTC (10,134 KB)
Full text under the publisher's open licence · 238 words
Related coverage
Multi-layer MIMO Relay as Deep Physical Neural Networks: Power Amplifiers as Activation Functions
arXiv · Jul 23, 2026
Data-driven Koopman mode approximation: A neural power iteration algorithm
arXiv · Aug 28, 2026
Researchers Find No Quantum Advantage in Time-Series Forecasting Models
Quantum Zeitgeist · Jul 29, 2026
Quantum Machine-Learning Design Targets Training And Scaling Barriers
The Quantum Insider · Jul 28, 2026