aiqbench analysis
The paper proposes making the routing between input signals and physical device variables a trainable parameter, optimized with Gumbel-Softmax, in broken-isomorphism physical neural networks. On MNIST and CIFAR-10 it reports accuracy gains (94.5% to 96.9% and 62.8% to 68.0%) using experimentally calibrated models, with little added inference cost. These are simulation-based results, so the practical question is whether the gains hold on fabricated hardware. Still, the work matters to hardware-oriented ML readers because it shows that learned wiring can extract more expressivity from physical substrates.
Optimizing Physical Neural Networks through Device Variable Selection
arXiv:2610.33522v1 Announce Type: cross Abstract: Physical neural networks (PNNs) leverage device physics in electronics, optics, and related domains to enable energy-efficient neural networks. Broken-isomorphism PNNs replace linear operations with multivariate physical transformations, modifying how input signals are mapped to device variables. We exploit network architectures emerging from these mappings by treating the routing between signals and input variables of physical devices as trainable parameters. This approach expands device-level tunability of the physical transformations while, at the system level, demonstrating improved expressivity of PNNs with negligible additional inference overhead. This discrete routing is optimized by training a categorical distribution using Gumbel-Softmax reparametrization. We evaluate this approach using experimentally calibrated differentiable models of multivariate electronic RNPUs for MNIST and CIFAR-10 classification. Using variable selection, we improve the MNIST classification accuracy for a small CNN model from 94.5% to 96.9% and the accuracy of CIFAR-10 from 62.8% to 68.0%. Our method uniquely leverages the induced biases of broken-isomorphism PNN architectures, paving the way towards more efficient physics-based neural network inference.
Abstract:Physical neural networks (PNNs) leverage device physics in electronics, optics, and related domains to enable energy-efficient neural networks. Broken-isomorphism PNNs replace linear operations with multivariate physical transformations, modifying how input signals are mapped to device variables. We exploit network architectures emerging from these mappings by treating the routing between signals and input variables of physical devices as trainable parameters. This approach expands device-level tunability of the physical transformations while, at the system level, demonstrating improved expressivity of PNNs with negligible additional inference overhead. This discrete routing is optimized by training a categorical distribution using Gumbel-Softmax reparametrization. We evaluate this approach using experimentally calibrated differentiable models of multivariate electronic RNPUs for MNIST and CIFAR-10 classification. Using variable selection, we improve the MNIST classification accuracy for a small CNN model from 94.5% to 96.9% and the accuracy of CIFAR-10 from 62.8% to 68.0%. Our method uniquely leverages the induced biases of broken-isomorphism PNN architectures, paving the way towards more efficient physics-based neural network inference.
Emerging Technologies (cs.ET); Hardware Architecture (cs.AR); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2610.33522 [cs.ET]
(or arXiv:2610.33522v1 [cs.ET] for this version)
https://doi.org/10.48550/arXiv.2610.33522 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 268 words
Related coverage
OptiSelect: How does the Optimizer Shape Data Curriculum?
arXiv · Oct 6, 2026
Agentic AI Reasoning and Learned Sampling for Task-Aware Projection Selection in Sparse-View CT: Predicting Hemodynamics
arXiv · Oct 31, 2026
First-order Constrained Trilevel Optimization Over Distributed Networks for Robust Coreset Selection
arXiv · Jul 31, 2026
Launching UI for generative AI inference recommendations in Amazon SageMaker AI
AWS · Jul 13, 2026