Compact Model Research Articles

A Compact Model for the Clustered Orienteering Problem

A Compact Model for the Clustered Orienteering Problem

Vision-aided grasp classification: design and evaluation of compact CNN for prosthetic hands.

Powered prosthetic hands capable of executing various grasp patterns are highly sought-after solutions for upper limb amputees. A crucial requirement for such prosthetic hands is the accurate identification of the intended grasp pattern and subsequent activation of the prosthetic digits accordingly. Vision-based grasp classification techniques offer improved coordination between amputees and prosthetic hands without physical contact. Deep learning methods, particularly Convolutional Neural Networks (CNNs), are utilized to process visual information for classification. The key challenge lies in developing a model that can effectively generalize across various object shapes and accurately classify grasp classes. To address this, a compact CNN model named GraspCNet is proposed, specifically designed for grasp classification in prosthetic hands. The use of separable convolutions reduces the computational burden, making it potentially suitable for real-time applications on embedded systems. The GraspCNet model is designed to learn and generalize from object shapes, allowing it to effectively classify unseen objects beyond those included in the training dataset. The proposed model was trained and tested using various standard object data sets. A cross-validation strategy has been adopted to perform better in seen and unseen object class scenarios. The average accuracy achieved was 82.22% and 75.48% in the case of seen, and unseen object classes respectively. In computer-based real-time experiments, the GraspCNet model achieved an accuracy of 69%. A comparative analysis with state-of-the-art techniques revealed that the proposed GraspCNet model outperformed most benchmark techniques and demonstrated comparable performance with the DcnnGrasp method. The compact nature of the GraspCNet model suggests its potential for integration with other sensing modalities in prosthetic hands.

Pruning Self-Attentions Into Convolutional Layers in Single Path.

Vision Transformers (ViTs) have achieved impressive performance over various computer vision tasks. However, modeling global correlations with multi-head self-attention (MSA) layers leads to two widely recognized issues: the massive computational resource consumption and the lack of intrinsic inductive bias for modeling local visual patterns. To solve both issues, we devise a simple yet effective method named Single-Path Vision Transformer pruning (SPViT), to efficiently and automatically compress the pre-trained ViTs into compact models with proper locality added. Specifically, we first propose a novel weight-sharing scheme between MSA and convolutional operations, delivering a single-path space to encode all candidate operations. In this way, we cast the operation search problem as finding which subset of parameters to use in each MSA layer, which significantly reduces the computational cost and optimization difficulty, and the convolution kernels can be well initialized using pre-trained MSA parameters. Relying on the single-path space, we introduce learnable binary gates to encode the operation choices in MSA layers. Similarly, we further employ learnable gates to encode the fine-grained MLP expansion ratios of FFN layers. In this way, our SPViT optimizes the learnable gates to automatically explore from a vast and unified search space and flexibly adjust the MSA-FFN pruning proportions for each individual dense model. We conduct extensive experiments on two representative ViTs showing that our SPViT achieves a new SOTA for pruning on ImageNet-1 k. For example, our SPViT can trim 52.0% FLOPs for DeiT-B and get an impressive 0.6% top-1 accuracy gain simultaneously.

Comparison of point cloud and image-based models for calorimeter fast simulation

Score based generative models are a new class of generative models that have been shown to accurately generate high dimensional calorimeter datasets. Recent advances in generative models have used images with 3D voxels to represent and model complex calorimeter showers. Point clouds, however, are likely a more natural representation of calorimeter showers, particularly in calorimeters with high granularity. Point clouds preserve all of the information of the original simulation, more naturally deal with sparse datasets, and can be implemented with more compact models and data files. In this work, two state-of-the-art score based models are trained on the same set of calorimeter simulation and directly compared.

End-to-end automated speech recognition using a character based small scale transformer architecture

This study explores the feasibility of constructing a small-scale speech recognition system capable of competing with larger, modern automated speech recognition (ASR) systems in both performance and word error rate (WER). Our central hypothesis posits that a compact transformer-based ASR model can yield comparable results, specifically in terms of WER, to traditional ASR models while challenging contemporary ASR systems that boast significantly larger computational sizes. The aim is to extend ASR capabilities to under-resourced languages with limited corpora, catering to scenarios where practitioners face constraints in both data availability and computational resources. The model, comprising a compact convolutional neural network (CNN) and transformer architecture with 2.214 million parameters, challenges the conventional wisdom that large-scale transformer-based ASR systems are essential for achieving high accuracy. In comparison, contemporary ASR systems often deploy over 300 million parameters. Trained on a modest dataset of approximately 3000 h—significantly less than the 50,000 h used in larger systems—the proposed model leverages the Common Voice and LibriSpeech datasets. Evaluation on the LibriSpeech test-clean and test-other datasets produced character error rates (CERs) of 6.40% and 16.73% and WERs of 16.03% and 35.51% respectively. Comparisons with existing architectures showcase the efficiency of our model. A gated recurrent unit (GRU) architecture, albeit achieving lower error rates, incurred a computational cost 24 times larger than our proposed model. Large-scale transformer architectures, while achieving marginally lower WERs (2%–4% on LibriSpeech test-clean), require 200 times more parameters and 53,000 additional hours of training data. Modern large language models are used to improve the WERs, but require large computational resources. To further enhance performance, a small 4-g language model was integrated into our end-to-end ASR model, resulting in improved WERs. The overarching goal of this work is to provide a practical solution for practitioners dealing with limited datasets and computational resources, particularly in the context of under-resourced languages.

Dynamic Characterization of SynRM With Dual-Axis Hybrid Excitation Self-Commissioning

The synchronous reluctance machine (SynRM) has gained increasing attraction for future electric drive systems. The characterization of its magnetic and current dynamics is essential for high-performance model-driven control, which still remains a promising problem. In this paper, a dynamic characterization scheme for SynRM with dual axis hybrid excitation self-comissioning is proposed to characterize the magnetic and current dynamics. Inspired by separating dual-axis current variables, a compact magnetic circuit model is investigated with reciprocity analytics. Furthermore, an improved current dynamic model is derived with resulting analytical apparent and incremental inductances. On this basis, the characterized parameters involved in the magnetic circuit and current dynamic models can be accurately identified by combining the designed dual axis hybrid excitation method with effective parameter identification algorithm. The effectiveness and practicability of the proposed scheme are verified based on the development platform for commercial SynRM drive. This simple dynamic characterization scheme fits for online implementation on standard drives. Compared with the existed work, 38.2 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> improvement of normalized modeling accuracy for magnetic circuit can be achieved. In the meanwhile, the normalized modeling errors of apparent inductances for current dynamics are strictly less than 6.91 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> .

DR-BERT: A protein language model to annotate disordered regions.

Despite their lack of a rigid structure, intrinsically disordered regions (IDRs) in proteins play important roles in cellular functions, including mediating protein-protein interactions. Therefore, it is important to computationally annotate IDRs with high accuracy. In this study, we present Disordered Region prediction using Bidirectional Encoder Representations from Transformers (DR-BERT), a compact protein language model. Unlike most popular tools, DR-BERT is pretrained on unannotated proteins and trained to predict IDRs without relying on explicit evolutionary or biophysical data. Despite this, DR-BERT demonstrates significant improvement over existing methods on the Critical Assessment of protein Intrinsic Disorder (CAID) evaluation datasetand outperforms competitors on two out of four test cases in the CAID 2 dataset, while maintaining competitiveness in the others. This performance is due to the information learned during pretraining and DR-BERT's ability to use contextual information.

PSP-Equivalent Model for Double-Gate and Surrounding-Gate Field Effect Transistors for Circuit Simulation

We introduce a compact core model for double-gate (DGFET) and surrounding-gate (SGFET) MOSFETs designed for circuit simulations. Despite its high precision, the model is crafted to retain the same analytic formulation of the industry standard Pennsylvania State and Philips (PSP). Instead of linearizing the drain current as in the PSP model, we employ a quadratic symmetric polynomial interpolation of the charge in the channel. This eliminates the need for cumbersome derivatives, simplifications, and intricate coding when integrating into a circuit simulator, thereby preventing singularities during numerical iterations. Moreover, thanks to its mathematical formulation equivalent to PSP, this model simplifies the coding of terminal charges, capacitances, potentials, and electric fields in the channel within circuit simulators. We validate the accuracy of the model through comparisons with numerical solutions and experiments from the literature.

Unique Relationship Between Optimum Compaction Properties of Fine-Grained Soils Across Rational Compactive Efforts: A Validation Study

AbstractAmong the many proposals for estimating the compaction characteristics of fine-grained soils for different compaction energy levels (CELs), energy conversion (EC) models are gaining increased attention. These models work on the premise of employing measured optimum moisture content (OMC) and maximum dry unit weight (MDUW) values obtained for a rational CEL (e.g., standard or reduced-standard Proctor (SP or RSP)) to predict the same for other CELs. This study revisits the most recently proposed EC-based compaction modeling framework, critically examining its asserted accuracy and hence identifying its true potentials. This was achieved by performing comprehensive statistical analyses on a newly compiled database of 206 compaction test results, entailing 70 different fine-grained soils (with liquid limits ranging 19–365%) and accounting for CELs of 202.0–2723.5 kJ/m3. It was demonstrated that 99% and 96% of the differences between the SP-converted and measured values for OMC and MDUW, respectively, fall within the allowable margins of OMC and MDUW measurement errors permitted by the Australian AS 1289.5.1.1/AS 1289.5.2.1 standards (satisfying their recommended ≥ 95% requirement). Equally favorable results were also obtained for the RSP-based conversions. These findings reaffirmed that the optimum compaction parameters across rational CELs are somewhat uniquely related, and the effects of fine-grained soil attributes on soil compactability are adequately captured/explained by the measured OMC and MDUW values employed as the conversion inputs/predictors.

Thin-Film Composite Membrane Compaction: Exploring the Interplay among Support Compressive Modulus, Structural Characteristics, and Overall Transport Efficiency.

Water scarcity has driven the demand for water production from unconventional sources and the reuse of industrial wastewater. Pressure-driven membranes, notably thin-film composite (TFC) membranes, stand as energy-efficient alternatives to the water scarcity challenge and various wastewater treatments. While pressure drives solvent movement, it concurrently triggers membrane compaction and flux deterioration. This necessitates a profound comprehension of the intricate interplay among compressive modulus, structural properties, and transport efficacy amid the compaction process. In this study, we present an all-encompassing compaction model for TFC membranes, applying authentic structural and mechanical variables, achieved by coupling viscoelasticity with Monte Carlo flux calculations based on the resistance-in-series model. Through validation against experimental data for multiple commercial membranes, we evaluated the influence of diverse physical parameters. We find that support polymers with a higher compressive modulus (lower compliance), supports with higher densities of "finger-like" pores, and "sponge-like" pores with optimum void fractions will be preferred to mitigate compaction. More importantly, we uncover a trade-off correlation between steady-state permeability and the modulus for identical support polymers displaying varying porosities. This model holds the potential as a valuable guide in shaping the design and optimization for further TFC applications and extending its utility to biological scaffolds and hydrogels with thin-film coatings in tissue engineering.

Universal Synaptic Plasticity of Interface‐Based Second‐Order Memristors

AbstractSecond‐order memristors with their internal short‐term dynamics display behavioral similarities with biological neurons and constitute an ideal basis unit for hardware neuromorphic networks, aims at treating spatio‐temporal tasks. Here, La0.7Sr0.3MnO3/BaTiO3/La0.7Sr0.3MnO3 second‐order memristive devices are investigated whose resistances and temperature dependencies range, on the same chip, from semiconductor to metal, but exhibit a universal neuromorphic plasticity. All devices may be described using a compact phenomenological model of current conduction, showing that resistive switching originates from interfaces, through charge trapping. Remarkably, the processes of short‐term memory gain/loss and long‐term consolidation/forgetting are the same whatever the device type. Only the synaptic transmission weights and the excitation/relaxation times with respect to stimuli differ, as it occurs for synapses/neurons in the brain. The weights may be tuned by the sole use of the frequency of stimuli (the activity rate), their evolution being dependent on previous activities (the history). Metal and semiconductor devices display the same in‐operando dynamics of potentiation or of depression, the transition from one regime to another being history‐dependent. The threshold frequencies are slightly lower in semiconducting devices. This work contributes to better understanding of memristive switching and plasticity and is relevant for the development of brain‐mimetic neural networks with new programming paradigms.

Kinetic models of HMX decomposition via chemical reaction neural network

1,3,5,7-Tetranitro-1,3,5,7-tetrazocane (HMX) is commonly used in solid propellants and explosives as energetic materials. The study of its kinetics and decomposition mechanism is of great significance to its application in the aerospace industries. This work investigates the thermal decomposition of HMX based on the combined thermogravimetric (TG) measurements and chemical reaction neural network (CRNN). Two compact kinetic models for HMX are introduced, with one consisting of four substances and a single global reaction (4–1 model) and the other consisting of four substances and four reactions (4–4 model). In the 4–1 model, the calculated activation energy is 328.44 kJ·mol−1, which agrees with the experimental value. As for the 4–4 model, the substances and reactions are assigned based on a skeleton mechanism involving reactions of N-N and C-N bond cleavage and HONO elimination. Moreover, the catalytic effects of TiO2 and Al2O3 on HMX are well simulated using the aforementioned kinetic models. The CRNN models can reproduce the peak temperature with a reduced activation energy, but the initial decomposition temperature is overestimated owing to the complex nature of catalytic impact. This work presents the application of the CRNN model to obtain a decomposition mechanism of HMX, highlighting its efficacy in accurately capturing the thermal decomposition behavior. The potential extension of CRNN to kinetic modeling of other energetic materials is anticipated in future studies.

Theoretical modeling and experimental investigation of in-phase resonant MEMS mirrors with cascaded structures

This paper proposes an efficient nonlinear one-dimensional (1D) compact mass-damper-spring model to predict the dynamic response of electrostatic resonant micro-electro-mechanical system (MEMS) mirrors with cascaded structures. The time-dependent damping moment due to viscous shear and pressure drag is computed using semi-empirical analytical equations for comb-drive structures and device frames. Nonlinear electrostatic force induced by the comb drives is efficiently acquired based on the hybrid method. The optimized device is fabricated using MEMS fabrication processes based on a 4-inch silicon-on-insulator wafer. The proposed compact model with the measured key parameters from the fabricated device shows excellent capability to accurately predict nonlinear dynamic responses of the fabricated device, including parametric excitation and hysteretic frequency response, with an average error of less than 5%. In particular, our 1D model is three orders of magnitude faster than the conventional finite element method model (0.8 s versus 1 h), enabling efficient system-level optimization of the critical design parameters. Based on the parametric study, electrode gap distance and torsion spring width are found to be two critical design parameters and dimensional analysis is conducted for design optimization with scan angle enhancing from 16.8° to 24° compared with the first design.

End-to-End Ultrasonic Hand Gesture Recognition

As the number of electronic gadgets in our daily lives is increasing and most of them require some kind of human interaction, this demands innovative, convenient input methods. There are limitations to state-of-the-art (SotA) ultrasound-based hand gesture recognition (HGR) systems in terms of robustness and accuracy. This research presents a novel machine learning (ML)-based end-to-end solution for hand gesture recognition with low-cost micro-electromechanical (MEMS) system ultrasonic transducers. In contrast to prior methods, our ML model processes the raw echo samples directly instead of using pre-processed data. Consequently, the processing flow presented in this work leaves it to the ML model to extract the important information from the echo data. The success of this approach is demonstrated as follows. Four MEMS ultrasonic transducers are placed in three different geometrical arrangements. For each arrangement, different types of ML models are optimized and benchmarked on datasets acquired with the presented custom hardware (HW): convolutional neural networks (CNNs), gated recurrent units (GRUs), long short-term memory (LSTM), vision transformer (ViT), and cross-attention multi-scale vision transformer (CrossViT). The three last-mentioned ML models reached more than 88% accuracy. The most important innovation described in this research paper is that we were able to demonstrate that little pre-processing is necessary to obtain high accuracy in ultrasonic HGR for several arrangements of cost-effective and low-power MEMS ultrasonic transducer arrays. Even the computationally intensive Fourier transform can be omitted. The presented approach is further compared to HGR systems using other sensor types such as vision, WiFi, radar, and state-of-the-art ultrasound-based HGR systems. Direct processing of the sensor signals by a compact model makes ultrasonic hand gesture recognition a true low-cost and power-efficient input method.

Employing post classification comparison to detect land use cover change patterns and quantify conversions in Abakaliki LGA Nigeria from 2000 to 2022

Rapid urbanization is restructuring landscapes across sub-Saharan Africa. This study employed post-classification comparison of multi-temporal Landsat imagery to characterize land cover changes in Abakaliki Local Government Area, Ebonyi State, Nigeria between 2000 and 2022, addressing the need for empirical baselines to guide sustainable planning. Four classes were considered and images classified with overall accuracy of 95% for the year 2000 and 97% for the year 2022. Notably, 21,000 hectares of vegetation were lost, while built-up and bare land increased by 7500 and 13,700 hectares respectively. Spatial patterns revealed built-up encroachment from vegetation and bare land; this establishes the first standardized quantification of Abakaliki LGA’s shifting landscape, with results supporting compact development models while conserving ecological services under ongoing transformations. The study makes a significant contribution by establishing an empirical baseline characterizing Nigeria's urbanization trajectory essential for evidence-based stewardship of regional resources and livelihoods in a period of accelerating change.

Deconfinement transitions in three-dimensional compact lattice Abelian Higgs models with multiple-charge scalar fields

Deconfinement transitions in three-dimensional compact lattice Abelian Higgs models with multiple-charge scalar fields

Pump-Driven Opto-Magnetic Properties in Semiconducting Transition-Metal Dichalcogenides: An Analytical Model.

Single-layer transition-metal dichalcogenides provide an unique intrinsic entanglement between the spin/valley/orbital degrees of freedom and the polarization of scattered photons. This scenario gives rise to the well-assessed optical dichroism observed by using both steady and time-resolved probes. In this paper, we provide compact analytical modeling of the onset of a finite Faraday/Kerr optical rotation upon shining with circularly polarized light. We identify different optical features displaying optical rotation at different characteristic energies, and we describe in an analytical framework the time-dependence of their intensities as a consequence of the main spin-conserving and spin-flip processes.

Accurate 2-D analytical model for cylindrical gate-junctionless ferroelectric-nanowire (CG-JFe-NW) MOSFET with scaled channel length

In this article, we propose a 2D analytical-model of a cylindrical gate-junctionless ferroelectric-nanowire (CG-JFe-NW) MOSFET for evaluating central potential, electric field, threshold voltage, mobility and current voltage characteristics. Central potential, threshold voltage, and current in the linear and saturation regions are derived by solving the 2D Poisson’s equation employing the parabolic-potential- approximation (PPA) profile. The recommended model is appropriate for compact modelling as it includes the phenomenon of bulk conduction, surface conduction and ferroelectric negative capacitance in all the regions of the device operation. Ferroelectric properties are introduced by combining Silicon doped hafnium oxide (ferroelectric layer) with HfO2 and SiO2 (dielectric layers), and an extensive study has been performed to determine the effects of interfacial-layer thickness and variation of channel length on device performance. For all relevant device characteristics and for all the operational voltage ranges, it is demonstrated that the analytical model entirely corresponds with the numerical simulations. For simulations, the Atlas 3-D simulator has been taken into consideration.

A Compact Model of Carbon Nanotube Field-Effect Transistors for Various Sizes with Bipolar Characteristics

Carbon nanotubes have excellent electrical properties and can be used as a new generation of semiconductor materials. This paper presents a compact model for carbon nanotube field-effect transistors (CNTFETs). The model uses a semi-empirical approach to model the current–voltage properties of CNTFETs with gate lengths exceeding 100 nm. This study introduces an innovative approach by proposing physical parametric reference lengths (Lref), which facilitate the integration of devices of varying sizes into a unified modeling framework. Furthermore, this paper develops models for the bipolar properties of carbon nanotube devices, employing two distinct sets of model parameters for enhanced accuracy. The model offers a comprehensive analysis of the different capacitances occurring between the electrodes within the device. The simulation of the model shows good agreement with the experimental measurements, confirming the model’s validity. The model is implemented in the Verilog-A hardware description language, with the circuit being subsequently constructed and subjected to simulations via the HSPICE tool. The CNTFET-based inverter exhibits a gain of 7.022 and a delay time of 16.23 ps when operated at a voltage of 1.2 V.

Experimentally realized physical-model-based frugal wave control in metasurface-programmable complex media

Metasurface-programmable radio environments are considered a key ingredient of next-generation wireless networks. Yet, identifying a metasurface configuration that yields a desired wireless functionality in an unknown complex environment was so far only achieved with closed-loop iterative feedback schemes. Here, we introduce open-loop wave control in metasurface-programmable complex media by estimating the parameters of a compact physics-based forward model. Our experiments demonstrate orders-of-magnitude advantages over deep-learning-based digital-twin benchmarks in terms of accuracy, compactness and required calibration examples. Strikingly, our parameter estimation also works without phase information and without providing measurements for all considered scattering coefficients. These unique generalization capabilities of our pure-physics model unlock unforeseen and previously inaccessible frugal wave control protocols that significantly alleviate the measurement complexity. For instance, we achieve coherent wave control (focusing or perfect absorption) and phase-shift-keying backscatter communications in metasurface-programmable complex media with intensity-only measurements. Our approach is also directly relevant to dynamic metasurface antennas, microwave-based signal processors and emerging in situ reconfigurable nanophotonic, optical and room-acoustical systems.