MNIST Data Research Articles

Leveraging Tunability of Localized-Interfacial Memristors for Efficient Handling of Complex Neural Networks.

A scalable (<130 nm) resistive switching memristor that features both filamentary and interfacial switching aimed at neuromorphic computing is developed in this study. The typically perceived noise or volatility was effectively harnessed as a controlled mechanism for interfacial switching. The multilayer structure for the proposed memristor enhances switching stability by curbing ionic overmigration and mitigating leakage paths. Furthermore, the memristors showcased their reliability by demonstrating more than 15 M cycles in the filamentary mode and 1 M pulses in the interfacial mode. Additionally, retention tests at 85 °C for 104 s confirmed the stability across different states, affirming its reliability as a nonvolatile CMOS-compatible element. While many studies validate performance solely on the MNIST data set, this work also evaluates more complex data sets, demonstrating the robustness of the demonstrated memristor in supervised learning. Specifically, supervised learning simulations on MNIST and fashion MNIST data sets indicated a high learning rate with <4% deviations from numerical training, while offline inference trained on CIFAR-10 and CIFAR-100 data sets revealed <2.5% and <7% deviations caused by programing error accumulation, even with increased memristor counts for these highly complex data sets. Unsupervised learning via spike-timing-dependent plasticity further highlights the potential of the developed memristor in bridging artificial and biological paradigms, offering a significant advance toward efficient and biologically inspired computing architectures.

Integrated Artificial Neural Network with Trainable Activation Function Enabled by Topological Insulator-Based Spin-Orbit Torque Devices.

Nonvolatile memristors offer a salient platform for artificial neural network (ANN), yet the integration of different function and algorithm blocks into one hardware system remains challenging. Here we demonstrate the brain-like synaptic (SOT-S) and neuronal (SOT-N) functions in the Bi2Te3/CrTe2 heterostructure-based spin-orbit torque (SOT) device. The SOT-S unit exhibits highly linear and symmetrical long-term potentiation/depression process, resulting in a fast-training of the MNIST data set with the classification accuracy above 90%. Meanwhile, the Sigmoid-shape transition curve inherited in the SOT-N cell replaces the software-based activation function block, hence reducing the system complexity. On this basis, we employ a serial-connected, voltage-mode sensing ANN architecture to enhance the vector-matrix multiplication signal strength with low reading error of 0.61% while simplifying the peripheral circuitry. Furthermore, the trainable activation function of SOT-N enables the implementation of the Batch Normalization algorithm and activation operation within one clock cycle, which bring about improved on/off-chip training performance close to the ideal baseline.

Reconfigurable Five-In-One Carbon Nanotube Optoelectronic Transistor for Intelligent Computing and Communication.

Advances in artificial general intelligence (AGI) necessitate the integration of diverse functionalities to address complex tasks. Carbon nanotubes (CNTs), with their unique physical properties, have broad applications in emerging research fields, providing a foundation for next-generation devices that could overcome the limits of Moore's Law. This work demonstrates a novel intelligent device that integrates five functions─sensors, memory, neuromorphic computing, logic, and communication─using CNT field-effect transistors (CNFETs) compatible with CMOS processes. Through passivation and annealing techniques, we have significantly enhanced the optoelectronic performance of CNFETs, leading to the development of multifunctional optoelectronic synaptic transistors. These optimized CNFETs enable dual-mode weight-tunable synaptic functions, including long-term plasticity and multilevel storage. Additionally, a CNT-based neural network has achieved high recognition accuracy on the MNIST data set, showcasing the potential of in-memory computing. This research also innovates by integrating logic functions with optoelectronic communication capabilities, paving the way for next-generation intelligent computing and communication integrated systems.

Trainable Reference Spikes Improve Temporal Information Processing of SNNs With Supervised Learning.

Spiking neural networks (SNNs) are the next-generation neural networks composed of biologically plausible neurons that communicate through trains of spikes. By modifying the plastic parameters of SNNs, including weights and time delays, SNNs can be trained to perform various AI tasks, although in general not at the same level of performance as typical artificial neural networks (ANNs). One possible solution to improve the performance of SNNs is to consider plastic parameters other than just weights and time delays drawn from the inherent complexity of the neural system of the brain, which may help SNNs improve their information processing ability and achieve brainlike functions. Here, we propose reference spikes as a new type of plastic parameters in a supervised learning scheme in SNNs. A neuron receives reference spikes through synapses providing reference information independent of input to help during learning, whose number of spikes and timings are trainable by error backpropagation. Theoretically, reference spikes improve the temporal information processing of SNNs by modulating the integration of incoming spikes at a detailed level. Through comparative computational experiments, we demonstrate using supervised learning that reference spikes improve the memory capacity of SNNs to map input spike patterns to target output spike patterns and increase classification accuracy on the MNIST, Fashion-MNIST, and SHD data sets, where both input and target output are temporally encoded. Our results demonstrate that applying reference spikes improves the performance of SNNs by enhancing their temporal information processing ability.

Image reconstruction through a nonlinear scattering medium via deep learning

Image reconstruction through the opaque medium has great significance in fields of biophotonics, optical imaging, mesoscopic physics, and optical communications. Previous researches are limited in the simple linear scattering process. Here, we develop a nonlinear speckle decoder network, which can reconstruct the phase information of the fundamental frequency wave via the nonlinear scattering signal. Further, we validate the ability of our model to recover simple and complex structures by using MNIST and CIFAR data sets, respectively. We then show that the model is able to restore the image information through different sets of nonlinear diffusers and reconstruct the image of a kind of completely unseen object category. The proposed method paves the way to nonlinear scattering imaging and information encryption.

Physics-Informed Masked Autoencoder for active sparse imaging

Imaging technology based on detecting individual photons has seen tremendous progress in recent years, with broad applications in autonomous driving, biomedical imaging, astronomical observation, and more. Comparing with conventional methods, however, it takes much longer time and relies on sparse and noisy photon-counting data to form an image. Here we introduce Physics-Informed Masked Autoencoder (PI-MAE) as a fast and efficient approach for data acquisition and image reconstruction through hardware implementation of the MAE (Masked Autoencoder). We examine its performance on a single-photon LiDAR system when trained on digitally masked MNIST data. Our results show that, with 1.8×10-6\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$1.8\ imes 10^{-6}$$\\end{document} or less detected photons per pulse and down to 9 detected photons per pixel, it achieves high-quality image reconstruction on unseen object classes with 90% physical masking. Our results highlight PI-MAE as a viable hardware accelerator for significantly improving the performance of single-photon imaging systems in photon-starving applications.

Perovskite-Doped Modulated Color-Selective Photosynaptic Transistors for Target Object Recognition.

The processing of multicolor noisy images in visual neuromorphic devices requires selective absorption at specific wavelengths; however, it is difficult to achieve this because the spectral absorption range of the device is affected by the type of material. Surprisingly, the absorption range of perovskite materials can be adjusted by doping. Herein, a CdCl2 co-doped CsPbBr3 nanocrystal-based photosensitive synaptic transistor (PST) is reported. By decreasing the doping concentration, the response of the PST to short-wavelength light is gradually enhanced, and even weak light of 40 μW·cm-2 can be detected. Benefiting from the excellent color selectivity of the PST device, the device array is applied to feature extraction of target blue items and removal of red and green noise, which results in the recognition accuracy of 95% for the noisy MNIST data set. This work provides new ideas for the application of novel transistors integrating sensors and storage computing.

Application of DMSFNN-COA technique for brand image design

Color plays a pivotal role in product design, as it can evoke emotional responses from users. Understanding these emotional needs is crucial for effective brand image design. This paper introduces a novel approach, the Brand Image Design using Deep Multi-Scale Fusion Neural Network optimized with Cheetah Optimization Algorithm (BID-DMSFNN-COA), for classifying product color brand images as “Stylish” and “Natural”. By leveraging deep learning techniques and optimization algorithms, the proposed method aims to enhance brand image accuracy and address existing challenges in product color trend forecasting research. Initially, data are collected from the Mnist Data Set. The data are then supplied into the pre-processing section. In the pre-processing segment, it removes the noise and enhances the input image utilizing master slave adaptive notch filter. The Deep Multi-Scale Fusion Neural Network optimized with cheetah optimization algorithm effectively classifies the product colour brand image as “Stylish” and “Natural”. Implemented on the MATLAB platform, the BID-DMSFNN-COA technique achieves remarkable accuracy rates of 99 % for both “Natural” and “Stylish” classifications. In comparison, existing methods such as BID-GNN, BID-ANN, and BID-CNN yield lower accuracy rates ranging from 65 % to 85 % for “Stylish” and 65 %–70 % for “Natural” product color brand image design. The simulation outcomes reveal the superior performance of the BID-DMSFNN-COA technique across various metrics including accuracy, F-score, precision, recall, sensitivity, specificity, and ROC analysis. Notably, the proposed method consistently outperforms existing approaches, providing higher values across all evaluation criteria. These findings underscore the effectiveness of the BID-DMSFNN-COA technique in enhancing brand image design through accurate product color classification.

Reservoir Computing Based on Oxygen-Vacancy-Mediated X-ray Optical Synaptic Device for Medical CT Bone Diagnosis.

Recognition and judgment of X-ray computed tomography (CT) images play a crucial role in medical diagnosis and disease prevention. However, the storage and calculation of the X-ray imaging system applied in the traditional CT diagnosis is separate, and the pathological judgment is based on doctors' experience, which will affect the timeliness and accuracy of decision-making. In this paper, a simple-structured reservoir computing network (RC) is proposed based on Ga2O3 X-ray optical synaptic devices to recognize medical skeletal CT images with high accuracy. Through oxygen vacancy engineering, Ga2O3 X-ray optical synaptic devices with adjustable photocurrent gain and a persistent photoconductivity effect were obtained. By using the Ga2O3 X-ray optical synaptic device as a reservoir, we constructed an RC network for medical skeletal CT diagnosis and verified its image recognition capability using the MNIST data set with an accuracy of 78.08%. In the elbow skeletal CT image recognition task, the recognition rate is as high as 100%. This work constructs a simple-structured RC network for X-ray image recognition, which is of great significance in applications in medical fields.

Synaptic Properties of a PbHfO3 Ferroelectric Memristor for Neuromorphic Computing.

The conventional von Neumann architecture has proven to be inadequate in keeping up with the rapid progress in artificial intelligence. Memristors have become the favored devices for simulating synaptic behavior and enabling neuromorphic computations to address challenges. An artificial synapse utilizing the perovskite structure PbHfO3 (PHO) has been created to tackle these concerns. By employing the sol-gel technique, a ferroelectric film composed of Au/PHO/FTO was created on FTO/glass for the purpose of this endeavor. The artificial synapse is composed of Au/PHO/FTO and exhibits learning and memory characteristics that are similar to those observed in biological neurons. The recognition accuracy for both MNIST and Fashion-MNIST data sets saw an increase, reaching 92.93% and 76.75%, respectively. This enhancement resulted from employing a convolutional neural network architecture and implementing an improved stochastic adaptive algorithm. The presented findings showcase a viable approach to achieve neuromorphic computation by employing artificial synapses fabricated with PHO.

Exploring deep echo state networks for image classification: a multi-reservoir approach

Echo state networks (ESNs) belong to the class of recurrent neural networks and have demonstrated robust performance in time series prediction tasks. In this study, we investigate the capability of different ESN architectures to capture spatial relationships in images without transforming them into temporal sequences. We begin with three pre-existing ESN-based architectures and enhance their design by incorporating multiple output layers, customising them for a classification task. Our investigation involves an examination of the behaviour of these modified networks, coupled with a comprehensive performance comparison against the baseline vanilla ESN architecture. Our experiments on the MNIST data set reveal that a network with multiple independent reservoirs working in parallel outperforms other ESN-based architectures for this task, achieving a classification accuracy of 98.43%. This improvement on the classical ESN architecture is accompanied by reduced training times. While the accuracy of ESN-based architectures lags behind that of convolutional neural network-based architectures, the significantly lower training times of ESNs with multiple reservoirs operating in parallel make them a compelling choice for learning spatial relationships in scenarios prioritising energy efficiency and rapid training. This multi-reservoir ESN architecture overcomes standard ESN limitations regarding memory requirements and training times for large networks, providing more accurate predictions than other ESN-based models. These findings contribute to a deeper understanding of the potential of ESNs as a tool for image classification.

Ballot Counting System Prototype with Hadoop Platform using Map and Reduce

Indonesia is among the nations practicing a democratic system, where the head of state is elected through general elections. During these elections, every citizen possessing a National Identity Card (KTP) holds equal voting rights. Counting votes accurately amidst the vast number of ballots poses a challenge. The ultimate objective is to develop a vote counting implementation using Hadoop, utilizing a form C1 image as input. The numerical data from the C1 form is processed into INT data types employing artificial neural networks (ANN) generated through the Restricted Boltzmann Machine (RBM), trained with MNIST data. These ANN models are uploaded onto a website for image storage in the database. The website's data is then crawled using Apache Nutch, and the obtained results are processed through Hadoop's MapReduce algorithm. Testing reveals efficient performance, with the recapitulation of results using Apache Nutch and Hadoop employing multimode and fair scheduler scheduling algorithms taking one minute and twenty-five seconds. Conversely, Hadoop configured with a single-node setup and FIFO scheduling algorithm achieves a quicker time of 67 seconds.

A Score-Based Deterministic Diffusion Algorithm with Smooth Scores for General Distributions

Score matching based diffusion has shown to achieve the state of art results in generation modeling. In the original score matching based diffusion algorithm, the forward equation is a differential equation for which the probability density equation evolves according to a linear partial differential equation, the Fokker-Planck equation. A drawback of this approach is that one needs the data distribution to have a Lipschitz logarithmic gradient. This excludes a large class of data distributions that have a compact support. We present a deterministic diffusion process for which the vector fields are always Lipschitz and hence the score does not explode for probability measures with compact support. This deterministic diffusion process can be seen as a regularization of the porous media equation equation, which enables one to guarantee long term convergence of the forward process to the noise distribution. Though the porous media equation is itself not always guaranteed to have a Lipschitz vector field, it can be used to understand the closeness of the output of the algorithm to the data distribution as a function of the the time horizon and score matching error. This analysis enables us to show that the algorithm has better dependence on the score matching error than approaches based on stochastic diffusions. Using numerical experiments we verify our theoretical results on example one and two dimensional data distributions which are compactly supported. Additionally, we validate the approach on a modified MNIST data set for which the distribution is concentrated on a compact set. In each of the experiments, the approach using deterministic diffusion performs better that the diffusion algorithm with stochastic forward process, when considering the FID scores of the generated samples.

Anomalous hall and skyrmion topological hall resistivity in magnetic heterostructures for the neuromorphic computing applications

Topologically protected spin textures, such as magnetic skyrmions, have shown the potential for high-density data storage and energy-efficient computing applications owing to their particle-like behavior, small size, and low driving current requirements. Evaluating the writing and reading of the skyrmion’s magnetic and electrical characteristics is crucial to implementing these devices. In this paper, we present the magnetic heterostructure Hall bar device and study the anomalous Hall and topological Hall signals in these devices. Using different measurement techniques, we investigate the magnetic and electrical characteristics of the magnetic structure. We measure the skyrmion topological resistivity and the magnetic field at different temperatures. MFM imaging and micromagnetic simulations further explain the anomalous Hall and topological Hall resistivity characteristics at various magnetic fields and temperatures. The study is extended to propose a skyrmion-based synaptic device showing spin-orbit torque-controlled plasticity. The resistance states are read using the anomalous Hall measurement technique. The device integration in a neuromorphic circuit is simulated in a 3-layer feedforward artificial neural network ANN. Based on the proposed synapses, the neural network is trained and tested on the MNIST data set, where a recognition accuracy performance of about 90% is achieved. Considering the nanosecond reading/writing time scale and a good system level performance, these devices exhibit a substantial prospect for energy-efficient neuromorphic computing.

Task-Adaptive Neuromorphic Computing Using Reconfigurable Organic Neuristors with Tunable Plasticity and Logic-in-Memory Operations.

The brain's function can be dynamically reconfigured through a unified neuron-synapse architecture, enabling task-adaptive network-level topology for energy-efficient learning and inferencing. Here, we demonstrate an organic neuristor utilizing a ferroelectric-electrolyte dielectric interface. This neuristor enables tunable short- to long-term plasticity and reconfigurable logic-in-memory functions by controlling the interfacial interaction between electrolyte ions and ferroelectric dipoles. Notably, the short-term plasticity of the organic neuristor allows for power-efficient reservoir computing in edge-computing scenarios, exhibiting impressive recognition accuracy, including images (90.6%) and acoustic signals (97.7%). For high-performance computing tasks, the neuristor based on long-term plasticity and logic-in-memory operations can construct all of the hardware circuits of a binarized neural network (BNN) within a unified framework. The BNN demonstrates excellent noise tolerance, achieving high recognition accuracies of 99.2% and 86.4% on the MNIST and CIFAR-10 data sets, respectively. Consequently, our research sheds light on the development of power-efficient artificial intelligence systems.

The Image Classification Algorithm Was Implemented on The MNIST Data Set

In the current era of rapid development of science and technology, the recognition and classification of digital images is the key to solving many problems, such as the application of license plate recognition, document digitization, and remote sensing image surface classification. Based on the MNIST handwritten numerical data set collected by the National Institute of Standards and Technology (NIST), this report uses Python language and PyTorch programming framework to construct a convolutional neural network (CNN) structure and practice and experience the image classification of handwritten digits in the MNIST data set.

Time regularization in optimal time variable learning

AbstractRecently, optimal time variable learning in deep neural networks was introduced in Antil et al. In this manuscript we extend the concept by introducing a regularization term that directly relates to the time horizon in discrete dynamical systems. Furthermore, we propose an adaptive pruning approach for Residual Neural Networks (ResNets), which reduces network complexity without compromising expressiveness, while simultaneously decreasing training time. The results are illustrated by applying the proposed concepts to classification tasks on the well known MNIST and Fashion MNIST data sets. Our PyTorch code is available on https://github.com/frederikkoehne/time_variable_learning.

QI$$^2$$: an interactive tool for data quality assurance

AbstractThe importance of high data quality is increasing with the growing impact and distribution of ML systems and big data. Also, the planned AI Act from the European commission defines challenging legal requirements for data quality especially for the market introduction of safety relevant ML systems. In this paper, we introduce a novel approach that supports the data quality assurance process of multiple data quality aspects. This approach enables the verification of quantitative data quality requirements. The concept and benefits are introduced and explained on small example data sets. How the method is applied is demonstrated on the well-known MNIST data set based an handwritten digits.

Monolithically Integrated Complementary Ferroelectric FET XNOR Synapse for the Binary Neural Network.

Neuromorphic computing, which mimics the structure and principles of the human brain, has the potential to facilitate the hardware implementation of next-generation artificial intelligence systems and process large amounts of data with very low power consumption. Among them, the XNOR synapse-based Binary Neural Network (BNN) has been attracting attention due to its compact neural network parameter size and low hardware cost. The previous XNOR synapse has drawbacks, such as a trade-off between cell density and accuracy. In this work, we show nonvolatile XNOR synapses with high density and accuracy using a monolithically stacked complementary ferroelectric field-effect transistor (C-FeFET) composed of a p-type Si MFMIS-FeFET at the bottom and a 3D stackable n-type Al:IZTO MFS-FeTFT, achieving 60F2 per cell (2C-FeFET). For adjusting the threshold voltage and improving the switching speed (100 ns) of n-type ferroelectric TFT, we employed a dual-gate configuration and a unique operation scheme, making it comparable to those of Si-based FeFETs. We performed array-level simulation with a 512 × 512 subarray size and a 3-bit flash ADC, demonstrating that the image recognition accuracies using the MNIST and CIFAR-10 data sets were increased by 3.17 and 14.07%, respectively, in comparison to other nonvolatile XNOR synapses. In addition, we performed system-level analysis on a 512 × 512 XNOR C-FeFET, exhibiting an outstanding throughput of 717.37 GOPS and an energy efficiency of 196.7 TOPS/W. We expect that our approach would contribute to the high-density memory systems, logic-in-memory technology, and hardware implementation of neural networks.

A triradical-containing trinuclear Pd(II) complex: spin-polarized electronic transmission, analog resistive switching and neuromorphic advancements.

Neuromorphic computation has emerged as a potential alternative to subvert the von Neumann bottleneck issue in conventional computing. In this context, the development of resistive switching-based memristor devices mimicking various synaptic functionalities has engendered paramount attention. Here, we report a triradical-containing trinuclear Pd(II) cluster with a cyclohexane-like framework constituted by the Pd-Se coordination motif displaying facile memristor property with neuromorphic functionality as a thin-film device. The metal-ligand complex (complex 1) possessed an St = 1/2 ground state by experiencing a spin-frustrated-type magnetic coupling phenomenon amongst the three ligand-based organic radicals (SR = 1/2), coordinated to the Pd(II) ions. Three reversible one-electron reduction waves countered with a one-electron and one two-electron reversible oxidation waves were noticed in the cyclic voltammogram of the complex, confirming electrons accepting and releasing capacity of the complex at low potentials, i.e., within +0.2 V to -1.1 V. Employing the radical-containing complex 1 as the active thin-film sandwiched between two orthogonal electrodes, resistive switching based memristor property with biological synaptic actions were successfully emulated. Intriguingly, the artificial neural network (ANN) simulated efficient pattern recognition demonstrated using the recorded potentiation and depression curves from the device, which is a step ahead for the hardware realization of neuromorphic computing. The performance of the ANN on MNIST data with reduced image resolution has further been evaluated. Density functional theory (DFT)-based theoretical calculation predicted that the spin-polarized electronic transmission substantiated the memristive property in the neutral complex 1.