One of the big challenges of current electronics is the design andimplementation of hardware neural networks that perform fast andenergy-efficient machine learning. Spintronics is a promising catalyst for thisfield with the capabilities of nanosecond operation and compatibility withexisting microelectronics. Considering large-scale, viable neuromorphic systemshowever, variability of device properties is a serious concern. In this paper,we show an autonomously operating circuit that performs hardware-aware machinelearning utilizing probabilistic neurons built with stochastic magnetic tunneljunctions. We show that $in \ situ$ learning of weights and biases in aBoltzmann machine can counter device-to-device variations and learn theprobability distribution of meaningful operations such as a full adder. Thisscalable autonomously operating learning circuit using spintronics-basedneurons could be especially of interest for standalone artificial-intelligencedevices capable of fast and efficient learning at the edge.
The search of hardware-compatible strategies for solving NP-hardcombinatorial optimization problems (COPs) is an important challenge of today scomputing research because of their wide range of applications in real worldoptimization problems. Here, we introduce an unconventional scalable approachto face maximum satisfiability problems (Max-SAT) which combines probabilisticcomputing with p-bits, parallel tempering, and the concept of invertible logicgates. We theoretically show the spintronic implementation of this approachbased on a coupled set of Landau-Lifshitz-Gilbert equations, showing apotential path for energy efficient and very fast (p-bits exhibiting ns timescale switching) architecture for the solution of COPs. The algorithm isbenchmarked with hard Max-SAT instances from the 2016 Max-SAT competition(e.g., HG-4SAT-V150-C1350-1.cnf which can be described with 2851 p-bits),including weighted Max-SAT and Max-Cut problems.
This paper proposes a method to provide a good initialization of control parameters to be found when performing manual or automated control tuning during development, commissioning or periodic retuning. The method is based on treating the initialization problem as a supervised learning one; taking examples from similar machines and similar tasks for which good control parameters have been found, and using those examples to build models that predict good control parameters for new machines and tasks yet to be initialized. Two of such models are proposed, one based on random forest regressors and a second based on neural networks. The random forest is highly data-efficient but generalizes only moderately. The neural network is able to leverage a high-dimensional burner run input to perform automatic system identification and generalization. While the proposed approach can be applied to a variety of applications for which example data from well functioning controllers can be used to hot-start new ones, we applied it in this paper to three slider-crank setups performing a variety of similar tasks. We found that both models outperform a benchmark of using a physics-inspired model for the initialization. Using 20% of the data for training, the required number of experiments was reduced up to 44%, and the performance of the initial experiments was improved by up to 68% compared to the benchmark.
Spintronic devices provide a promising beyond-complementary metal-oxide-semiconductor (CMOS) device option, thanks to their energy efficiency and compatibility with CMOS. To accurately capture their multiphysics dynamics, a rigorous treatment of both spin and charge and their inter-conversion is required. Here, we present physics-based device models based on $4\times4$ matrices for the spin-orbit coupling (SOC) part of the magneto-electric spin-orbit (MESO) device. Also, a more rigorous physics model of ferroelectric and magnetoelectric (ME) switching of ferromagnets, based on Landau–Lifshitz–Gilbert (LLG) and Landau–Khalatnikov (LK) equations, are presented. With the combined model implemented in a SPICE circuit simulator environment, simulation results were obtained which show feasibility of the MESO implementation and the functional operation of buffers, synchronous oscillators, and majority gates.
The nearing end of Moore's Law has been driving the development of domain-specific hardware tailored to solve a special set of problems. Along these lines, probabilistic computing with inherently stochastic building blocks (p-bits) have shown significant promise, particularly in the context of hard optimization and statistical sampling problems. p-bits have been proposed and demonstrated in different hardware substrates ranging from small-scale stochastic magnetic tunnel junctions (sMTJs) in asynchronous architectures to large-scale CMOS in synchronous architectures. Here, we design and implement a truly asynchronous and medium-scale p-computer (with $\approx \mathbf{800}\ \mathbf{p}-\mathbf{bits}$) that closely emulates the asynchronous dynamics of sMTJs in Field Programmable Gate Arrays (FP-GAs). Using hard instances of the planted Ising glass problem on the Chimera lattice, we evaluate the performance of the asynchronous architecture against an ideal, synchronous design that performs parallelized (chromatic) exact Gibbs sampling. We find that despite the lack of any careful synchronization, the asynchronous design achieves parallelism with comparable algorithmic scaling in the ideal, carefully tuned and parallelized synchronous design. Our results highlight the promise of massively scaled p-computers with millions of free-running p-bits made out of nanoscale building blocks such as stochastic magnetic tunnel junctions.
The development of accurate methods for multi-label classification (MLC) of remote sensing (RS) images is one of the most important research topics in RS. The MLC methods based on convolutional neural networks (CNNs) have shown strong performance gains in RS. However, they usually require a high number of reliable training images annotated with multiple land-cover class labels. Collecting such data is time-consuming and costly. To address this problem, the publicly available thematic products, which can include noisy labels, can be used to annotate RS images with zero-labeling cost. However, multi-label noise (which can be associated with wrong and missing label annotations) can distort the learning process of the MLC methods. To address this problem, we propose a novel multi-label noise robust collaborative learning (RCML) method to alleviate the negative effects of multi-label noise during the training phase of a CNN model. RCML identifies, ranks, and excludes noisy multi-labels in RS images based on three main modules: 1) the discrepancy module; 2) the group lasso module; and 3) the swap module. The discrepancy module ensures that the two networks learn diverse features, while producing the same predictions. The task of the group lasso module is to detect the potentially noisy labels assigned to multi-labeled training images, while the swap module is devoted to exchange the ranking information between two networks. Unlike the existing methods that make assumptions about noise distribution, our proposed RCML does not make any prior assumption about the type of noise in the training set. The experiments conducted on two multi-label RS image archives confirm the robustness of the proposed RCML under extreme multi-label noise rates. Our code is publicly available at: https://www.noisy-labels-in-rs.org .
This commentary is on the original article by Danks et al. on pages 1517–1523 ofthis issue.