Sep 2026· IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems· Vol 45, pp. 4408-4421· 0 citations· 76 references
Abstract
This article introduces KirchhoffNet, a novel class of neural network models inspired by the principles of analog electronic circuitry, specifically Kirchhoff’s laws. KirchhoffNet operates as an analog circuit, where the network input is represented by initial node voltages, and the output corresponds to the node voltages at a specific time. The dynamics of the node voltages are governed by learnable parameters on the edges, and the evolution of these voltages follows a system of ordinary differential equations (ODEs). Despite the absence of traditional neural network components such as convolutional layers, KirchhoffNet achieves outstanding performance across a wide range of machine-learning tasks. We further demonstrate that KirchhoffNet is capable of computing diffusion models, making it a promising candidate for accelerating modern generative AI applications. Most notably, KirchhoffNet can be implemented as a high-speed & low-power analog integrated circuit, which introduces a compelling advantage: irrespective of the number of parameters in the network, its on-chip forward calculation can always be completed within a short time. This property makes KirchhoffNet a highly attractive and scalable paradigm for implementing large-scale neural networks, opening new avenues in the realm of analog neural networks for artificial intelligence (AI).
Oscillatory Neural Networks (ONNs) are an alternative computing paradigm for AI and combinatorial optimization problems. However, digital architectures are often designed for specific applications of ONNs. This work introduces a modular and scalable architecture called ODEONN that is generic to multiple applications of ONNs, and to the best of our knowledge, is the first fully digital ONN to also support complex-valued coupling. Additionally, an approximation of the sine function is introduced that uses half of the hardware resources compared to standard methods. The performance of ODEONN is compared with a full-precision software simulation, where a performance degradation of less than $2\%$ is shown. Therefore, we conclude that the fixed-point quantization and the approximated waveform affect the accuracy of computation by only a small amount. Furthermore, ODEONN shows a 45$\times$ reduction in energy-delay product over the software simulation running on conventional hardware.
Analog compute paradigms are gaining attention for their potential to overcome the energy and latency limits of digital systems. Analog computations are modeled as dynamical systems for design and optimization, but this process is challenging: the dynamics are governed by nonlinear differential equations without closed-form solutions, are susceptible to hardware nonidealities such as mismatch and noise, and rely on digitally programmable interfaces that complicate system-level reasoning. We present Shem , a framework that enables gradient-based optimization of system-level objectives such as robustness and accuracy on user-defined analog compute models. We leverage the fact that analog compute models are inherently differentiable, and that nonidealities and digital interfaces can be approximated by differentiable functions, making gradient descent a natural and unified approach for optimization. We evaluate Shem across diverse case studies, including an oscillator-based pattern recognizer, a cellular nonlinear network edge detector, an analog-to-information converter integrated with a neural network, and a transmission-line security primitive. In all cases, Shem improves application performance, including signal-to-noise ratio, classification accuracy, and a security metric. In addition, we introduce an autoparallelization algorithm that reduces total optimization runtime by 49–91%.
Yu-Neng Wang, Sara Achour· International Conference on...· 0 citations
It is aimed at proving that SNNs have potential in such areas as computer vision, robotics, and speech recognition, and their role in overcoming the barrier between artificial and biological neural systems is proved.
Mesala Sravani, K. Kumari, S. M. Reddy· International Journal of Unc...· 0 citations
This work proposes a noisy group neuron (NGN) model, which incorporates population-level synchronous resetting and neural stochasticity as fundamental computational mechanisms, and develops the NGN method as a framework that combines the NGN model with backpropagation learning based on mean-field dynamics.
Yajie Zhai, Yanmei Kang, Meng Li et al.· 0 citations
Hard-constrained recurrent physics-informed networks (HRPINNs) embed known dynamics inside a recurrent numerical integrator and restrict a neural branch to learning only the residual dynamics that the first-principles model does not capture. Kolmogorov--Arnold Networks (KANs) have been proposed as parameter-efficient replacements for multilayer perceptrons (MLPs) in such residual branches, but their learnable B-spline activations follow a markedly different execution profile. Building on prior work that characterized when a vanilla B-spline KAN matches or underperforms an MLP as an HRPINN residual branch in discovery accuracy, this paper asks whether that parameter efficiency survives deployment. Using identical trained weights, we measured execution latency, energy per integration step, and dependability under post-training quantization in the closed recurrent loop on a RISC-V RV64GC platform without vector extensions (StarFive VisionFive~2, SiFive U74). For the two accuracy-comparable pairs, the KAN residual branch executed $13.5\times$ and $8.0\times$ slower and consumed $11.3\times$ and $5.6\times$ more energy per integration step (3.7\,$\mu$J against 0.33\,$\mu$J for the smallest pair); across all four parameter-matched size tiers the ranges are $4.7\times$--$14.5\times$ and $4.7\times$--$18.7\times$. Under INT8 quantization, KAN trajectories diverged up to $43\times$ earlier than matched MLPs; the damage traces to weight quantization, not to input-side knot-interval misassignment. These results indicate that the parameter efficiency reported for KANs does not transfer to deployment cost on scalar embedded cores, and that an MLP residual branch is the more dependable default for embedded HRPINN deployment unless specific quantization co-design is used.
Backpropagation (BP) dominates deep learning training, but its reliance on gradients brings inherent troubles -- vanishing and exploding gradients. The pursuit of gradient-free methods has long been a goal in the field of artificial intelligence. This paper shows that indeed the simplest Monte Carlo algorithm implemented on a single GPU -- randomly mutate a parameter, keep it if the loss decreases, otherwise retry -- can practically train deep networks. This gradient-free method does not even need common techniques such as batch normalization or residual connections to directly train sufficiently deep networks. More remarkably, its flexibility extends to several nontrivial scenarios: it enables pure pruning training, supports discrete weights, accommodates unconventional transfer functions such as Gaussian, and reveals the substantial redundancy of deep networks. We have demonstrated its feasibility on deep networks with more than 20 layers, single-hidden-layer wide networks with up to 16,384 hidden neurons, and even a simple Transformer architecture trained on both image classification (MNIST) and character-level language modeling (Tiny Shakespeare). This simple gradient-free method may offer a complementary perspective for understanding the self-organization and learning mechanisms of neural networks, and also provides an alternative route for building physically inspired deep learning systems.
Hong Zhao· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduJul 14, 2026
Through research and entrepreneurship, Professor Devavrat Shah is helping to design methods that can handle constant decision-making using limited computational resources.