Computational Capabilities Research Articles

Evolutionary Computing Control Strategy of Nonholonomic Robots with Ordinary Differential Equation Kinematics Model

This paper introduces an Evolutionary Computing Control Strategy (ECCS) for the motion control of nonholonomic robots, and integrates an ordinary differential equation (ODE)-based kinematics model with a nonlinear model predictive control (NMPC) strategy and a particle-based evolutionary computing (PEC) algorithm. The ECCS addresses the key challenges of traditional NMPC controllers, such as their tendency to fall into local optima when solving nonlinear optimization problems, by leveraging the global optimization capabilities of evolutionary computation. Experiment results on the MATLAB Simulink platform demonstrate that the proposed ECCS significantly improves motion control accuracy and reduces control errors compared to linearized MPC (LMPC) strategies. Specifically, the ECCS reduces the maximum error by 90.6% and 94.5%, the mean square error by 67.8% and 92.6%, and the root mean square error by 43.5% and 70.3% in velocity control and steering angle control, respectively. Furthermore, experiments are separately implemented on the CarSim platform and the physical environment to verify the availability of the proposed ECCS. Furthermore, experiments are separately implemented on the CarSim platform and the physical environment to verify the availability of the proposed ECCS. These results validate the effectiveness of embedding ODE kinematics into the evolutionary computing framework for robust and efficient motion control of nonholonomic robots.

Detailed validation of LES for H[formula omitted]/CH[formula omitted]/Air deflagrations in an obstructed tube using PIV measurements

This study offers a detailed validation of Large Eddy Simulation (LES) for lean H2/CH4/Air deflagrations in an obstructed tube. An exhaustive validation is conducted against detailed measurements from Li et al. (2019), which include pressure traces, flame speeds, and especially, Particle Image Velocimetry measurements of the deflagration-induced flow field. The exercise is performed without adjusting any model parameters, so that all simulations are executed using a unique numerical setup across all test cases. This approach provides a robust and unbiased assessment of LES capabilities in capturing the complex interactions between flame propagation, turbulence, and obstacles in explosion scenarios. Results demonstrate that LES accurately predicts the detailed evolution of the flow field in the recirculation zone behind the second obstacle, and the resulting over-pressure as well as flame speed and flame qualitative shape for various deflagration severities. Such results highlight the potential of LES for improving Safety Computational Fluid Dynamics predictive capabilities in industrial applications involving explosive environments. Once validated, LES is analyzed to unravel flame propagation dynamics: It is demonstrated that the flame remains laminar-like up to the second obstacle and then transitions to the turbulent combustion regime. Independently from the mixture blend, the maximum over-pressure is correlated to flame-turbulence interactions occurring in the wake of the second obstacle. While LES effectively captures these dynamics, it is noted that usual methods to quantify flows in pipes are inadequate for fully characterizing the transition to turbulence: Developing more refined indicators to detect this transition are required.Novelty and significance statementThis study presents a significant advancement in the validation of Large Eddy Simulation (LES) for complex deflagration scenarios within obstructed geometries. Unlike previous works that typically rely on pressure data, flame speeds, and basic visualizations, this research integrates comparisons to Particle Image Velocimetry measurements for a quantitative validation of LES deflagrations in obstructed channels. By leveraging the detailed experimental dataset from Li et al. (2019), this paper establishes a new benchmark for simulation accuracy, demonstrating LES ability to capture complex flame-turbulence interactions in confined spaces. This work not only addresses the critical gap in the literature but also opens the way for advancements in Safety Computational Dynamics, setting a higher standard for future simulation studies.

Stoch-IMC: A bit-parallel stochastic in-memory computing architecture based on STT-MRAM

In-memory computing (IMC) offloads parts of the computations to memory to fulfill the performance and energy demands of applications such as neuromorphic computing, machine learning, and image processing. Fortunately, the main features that stochastic computing (SC) and IMC share, which are low computation complexity and high bit-parallel computation capability, promise great potential for integrating SC and IMC. In this paper, we exploit this potential by using stochastic computation as an approximation method to present effective in-memory computations with a good trade-off among design parameters. To this end, first, commonly used stochastic arithmetic operations of applications are effectively implemented using the primitive logic gates of the IMC method. Next, the in-memory scheduling and mapping of applications are obtained efficiently by a proposed algorithm. This algorithm reduces the computation latency by enabling intra-subarray parallelism while considering the IMC method constraints. Subsequently, a bit-parallel stochastic IMC architecture, Stoch-IMC, is presented that enables bit parallelization of stochastic computations over memory subarrays/banks. To evaluate Stoch-IMC’s effectiveness, various analyses were conducted. Results show average performance improvements of 135.7X and 124.2X across applications compared to binary IMC and related in-memory SC methods, respectively. The results also demonstrate an average energy reduction of 1.5X compared to binary IMC, with limited energy overhead relative to the in-memory SC method. Furthermore, the results reveal average lifetime improvements of 4.9X and 216.3X over binary IMC and in-memory SC methods, respectively, along with high bitflip tolerance.

Redundant task offloading with dual-reliability in MEC-assisted vehicular networks

With the rise and development of intelligent vehicles, the computation capability of vehicles has increased rapidly and considerably. Vehicle-to-Vehicle (V2V) offloading, in which computation-intensive tasks are offloaded to underutilized vehicles, has been proposed. However, V2V offloading faces the challenges of task transmission reliability and task computation reliability. In V2V offloading, tasks are transmitted via V2V communication, which is volatile and spotty because of rapidly changing network topology and channel conditions between vehicles, resulting in time-varying delays of task transmission and even loss of connectivity. Thus, it is challenging to complete V2V offloading within a given delay constraint. In addition, the realistic diverse vehicular environment always comes with malicious vehicles, which can cause irreparable harm to V2V offloading. Therefore, in this paper, we propose a V2V task offloading scheme called Redundant Task Offloading with Dual-Reliability (RTODR), aiming to minimize task offloading costs while ensuring both task transmission reliability and task computation reliability in a Mobile Edge Computing (MEC)-assisted vehicular network. Specifically, for a computation task, a V2V connection is considered reliable only if the task can be successfully transmitted via the V2V connection within the deadline of the task. To ensure task computation reliability, task computation results from a trusty service vehicle are considered to be reliable. Then we formally model a Minimizing Task Offloading Cost with Dual-reliability (MTOCD) problem, which is mathematically formulated as a multi-objective optimization problem. Afterward, we propose a heuristic redundant task offloading algorithm, named Dual-Reliability Offloading (DRO), to solve the problem. Finally, comprehensive experiments have been conducted to demonstrate that RTODR achieves lower costs compared with other approaches.

Hybrid Tree Tensor Networks for Quantum Simulation

Hybrid tensor networks (hTNs) offer a promising solution for encoding variational quantum states beyond the capabilities of efficient classical methods or noisy quantum computers alone. However, their practical usefulness and many operational aspects of hTN-based algorithms, like the optimization of hTNs, the generalization of standard contraction rules to an hybrid setting, and the design of application-oriented architectures have not been thoroughly investigated yet. In this work, we introduce a novel algorithm to perform ground-state optimizations with hybrid tree tensor networks (hTTNs), discussing its advantages and roadblocks, and identifying a set of promising applications. We benchmark our approach on two paradigmatic models, namely the Ising model at the critical point and the Toric-code Hamiltonian. In both cases, we successfully demonstrate that hTTNs can improve upon classical equivalents with equal bond dimension in the classical part. Published by the American Physical Society 2025

The Economic Impacts of Quantum Computing on Venture Capital Investment

Quantum computing offers transformative computational capabilities, disrupting traditional investment paradigms and reshaping venture capital (VC) strategies. This paper examines its economic impact on VC, highlighting advancements in cryptography, materials science, and AI that create opportunities for startups and alter risk-reward dynamics. By analyzing trends and case studies, we reveal how quantum innovation accelerates cycles, redefines valuations, and reshapes global competition, guiding investors, policymakers, and entrepreneurs in navigating the emerging quantum landscape.

Air-Writing Recognition using Deep learning and Artificial Intelligence

This paper investigates the use of deep learning architectures for recognizing 3D gestures, particularly focusing on air-writing, performed using hand or finger movements. Unlike traditional handwriting, which involves distinct pen-up and pen-down actions, air-writing lacks a defined sequence of such events. Additionally, the absence of visual or tactile feedback and the high dependency among target gestures pose significant challenges for accurate character recognition. From the user's perspective, air-writing can be executed in three distinct styles: connected, isolated, and overlapped. In the connected style, users write sequences of characters in free space, similar to writing on paper from left to right. The isolated approach confines each character’s gesture within an imaginary 3D box, making segmentation easier but less natural. The overlapped method is the most complex, as it involves stacking adjacent characters within the same imaginary space. Variations in character size and shape during connected writing can impact recognition accuracy, even for the same user, while the constrained isolated style offers more consistent segmentation. Advancements in Motion Capture (MoCap) technologies and increased computational capabilities have encouraged the exploration of air-writing in human-computer interaction (HCI) systems. Some systems utilize handheld tools or custom gloves for tracking motion trajectories, while low-cost depth sensors, such as Microsoft Kinect, Intel RealSense, and Leap Motion Controller (LMC), offer non-intrusive, real-time tracking solutions. Kinect focuses on long-range 3D posture detection, whereas LMC and RealSense provide millimeter-level precision for tracking hand and finger movements. However, distinguishing actual writing gestures from arbitrary hand movements in continuous motion streams remains a significant challenge, independent of character recognition. Existing air-writing systems primarily target letter and digit recognition due to the limited availability of word-level training datasets. To address this gap, we propose a solution capable of both detecting and recognizing air-written characters within continuous motion data. Our system, based on the LMC sensor, includes a web interface for capturing 1,200 air-written digits (0–9) while providing real-time visual feedback. An efficient fingertip tracking mechanism simulates pen-up and pen-down states. The air-writing recognition process is framed as a classification task, utilizing both dynamic 3D trajectories (time-series data) and static 2D stroke projections (images) to train deep learning models with convolutional and recurrent architectures. Experimental results demonstrate near-perfect recognition rates achieved within milliseconds, making the system suitable for real-time deployment. The paper is organized as follows: Section II reviews related works in air-writing. Section III outlines the methodology, including the LMC interface, dataset collection, and employed deep learning models. Section IV describes the experimental setup and evaluation results. Finally, Section V concludes with key findings and future research directions..

QuanFormer: A Transformer-Based Precise Peak Detection and Quantification Tool in LC-MS-Based Metabolomics.

In metabolomic analysis based on liquid chromatography coupled with mass spectrometry, detecting and quantifying intricate objects is a massive job. Current peak picking methods still cause high rates of incorrectly picked peaks to influence the reliability and reproducibility of results. To address these challenges, we developed QuanFormer, a deep learning method based on object detection designed to accurately quantify peak signals. Our algorithm combines the feature extraction capabilities of convolutional neural networks (CNNs) with the global computation capability of Transformer architecture. Data training in QuanFormer by using nearly 20,000 annotated regions-of-interest (ROIs) ensures unique prediction via bipartite matching, achieving 96.5% of the average precision value on the test set. Even without retraining, QuanFormer achieves over 90% accuracy in distinguishing true from false peaks. Performance was further analyzed using visualization techniques applied to the encoder and decoder layers. We also demonstrated that QuanFormer could correct retention time shifts for peak alignment and generally surpass the existing methods, including MZmine 3 and PeakDetective, to obtain a larger number of picked peaks and higher accurate quantification. Finally, we also carried out metabolomic analysis in a clinical cohort of breast cancer patients and utilized QuanFormer to detect and quantify the potential biomarkers. QuanFormer is open-source and available at https://github.com/LinShuhaiLAB/QuanFormer.

Introduction to Memristive Mechanisms and Models.

The increase in computational power demand led by the development of Artificial Intelligence is rapidly becoming unsustainable. New paradigms of computation, which potentially differ from digital computation, together with novel hardware architecture and devices, are anticipated to reduce the exorbitant energy demand for data-processing tasks. Memristive systems with resistive switching behavior are under intense research, given their prominent role in the fabrication of memory devices that promise the desired hardware revolution in our intensive data-driven era. They are suggested to provide the hardware substrate to scale up computational capabilities while improving their energy expenditure and speed. This work provides an orientation map for those interested in the vast topic of memristive systems with application to neuromorphic computing. We address the description of the most notable emerging devices and we illustrate models that capture the complex dynamical behavior of these systems under the dynamical-systems framework developed by Chua. We then review the memristive behavior under the perspective of statistical physics and percolation theory suited to describe fluctuations and disorder which are otherwise precluded in the dynamical-system approach. Percolation theory allows the investigation of these systems at the mesoscopic level, enabling material-independent modeling of non-linear conductance networks. We finally discuss recent and less recent successes in deep learning methods that bridge the field of physics-based and biological- inspired neuromorphic computing.

Large AI Models and Their Applications: Classification, Limitations, and Potential Solutions

ABSTRACTBackgroundIn recent years, Large Models (LMs) have been rapidly developed, including large language models, visual foundation models, and multimodal LMs. They are updated and iterated at a very fast pace. These LMs can accomplish many tasks, e.g., daily work assistant, intelligent customer service, and intelligent factory scheduling. Their development has contributed to various industries in human society.AimsThe architectural flaws of LMs lead to several problems, including illusions and difficulty in locating errors, limiting their performance. Solving these problems properly can facilitate their further development.MethodsThis work first introduces the development of LMs and identifies their current problems, including data and energy consumption, catastrophic forgetting, reasoning ability, localization fault, and ethical problems. Then, potential solutions to these problems are provided, including increase data and computation capability, neural‐symbolic synergy, and data orientation to human pattern.DiscussionThis work discusses developing vertical domain LMs on top of some base LMs. In addition, this work introduces three typical real‐world applications of LMs, including autonomous driving, smart industrial productions, and intelligent medical assistance.ConclusionBy embracing the advantages of LMs and solving their fundamental problems, many industries are expected to achieve promising prospects in the future.

Exploration of the Impact of Digital Labor on the Capital Production Process based on Big Data Algorithms

In the era of digital economy, the computing and storage capabilities of computers drive the rapid development of digital technologies such as big data algorithms. Digital capital relies on digital technology to establish digital enterprises, making digital labor a new form of labor. However, there is controversy in academia over whether digital labor creates value and surplus value. This article analyzes the impact of digital labor on capital production from the perspective of big data algorithms, and finds that although digital labor changes the premise of capital production, it still belongs to "productive labor" and can create value and surplus value. On the surface, digital enterprises provide free and flexible work, but in reality, they are conducive to capital production. By monitoring workers through big data algorithms, they ensure production efficiency and quality. This "zero cost" form of employment breaks the capital limit of increasing the number of employees and enables the rapid proliferation of digital capital. Therefore, digital labor has not changed the essence of capital production, but has made digital labor production appear "technologically invisible" through the use of digital technology.

Critical Factors in Parabolic Nozzle Design and Performance Analysis with CFD

Abstract This study presents a comprehensive analysis of contoured nozzle designs using the Method of Characteristics (MOC) and Fluent simulations. The study aims to compare the performance and flowfield predictions of these methods and explore the influence of various design param- eters on nozzle performance. Initial investigations validate the MOC algorithm’s accuracy against commercial software, highlighting its rapid computational capability despite limita- tions in handling real gas effects. Fluent simulations consistently predict slightly lower specific impulse values due to differing exit pressure predictions but align closely with MOC results in inviscid flow conditions. The research explores the role of the initial circular arc in defin- ing the nozzle’s expansion and straightening sections, finding that smaller circular arcs create larger property gradients, necessitating quicker expansions and leading to potential isentropic losses, which were almost imperceptible in the Fluent case, in fact, a more axial flow was deliv- ered. Variations in design parameters, such as the contour’s exit angle and cone nozzle fraction, demonstrate their impact on nozzle performance, with higher exit angles generally improving specific impulse, contrary to expectation. For some pairs of design options, longer nozzles al- low smoother expansions, resulting in lower average exit angles and improved specific impulse. The findings emphasize the contemporary relevance of the MOC for preliminary nozzle studies, highlighting the balance between accuracy and computational cost.

Harnessing ferro-valleytricity in pentalayer rhombohedral graphene for memory and compute

Two-dimensional materials with multiple degrees of freedom, including spin, valleys, and orbitals, open up an exciting avenue for engineering multifunctional devices. Beyond spintronics, these degrees of freedom can lead to novel quantum effects such as valley-dependent Hall effects and orbital magnetism, which could revolutionize next-generation electronics. However, achieving independent control over valley polarization and orbital magnetism has been a challenge due to the need for large electric fields. A recent breakthrough involving pentalayer rhombohedral graphene has demonstrated the ability to individually manipulate anomalous Hall signals and orbital magnetic hysteresis, forming what is known as a valley-magnetic quartet. Here, we leverage the electrically tunable ferro-valleytricity of pentalayer rhombohedral graphene to develop nonvolatile memory and in-memory computation applications. We propose an architecture for a dense, scalable, and selector-less nonvolatile memory array that harnesses the electrically tunable ferro-valleytricity. In our designed array architecture, nondestructive read and write operations are conducted by sensing the valley state through two different pairs of terminals, allowing for independent optimization of read/write peripheral circuits. The power consumption of our PRG-based array is remarkably low, with only ∼6 nW required per write operation and ∼2.3 nW per read operation per cell. This consumption is orders of magnitude lower than that of the majority of state-of-the-art cryogenic memories. Additionally, we engineer in-memory computation by implementing majority logic operations within our proposed nonvolatile memory array without modifying the peripheral circuitry. Our framework presents a promising pathway toward achieving ultra-dense cryogenic memory and in-memory computation capabilities.

Exploring In-Network Computing with Information-Centric Networking: Review and Research Opportunities

The advent of 6G networks and beyond calls for innovative paradigms to address the stringent demands of emerging applications, such as extended reality and autonomous vehicles, as well as technological frameworks like digital twin networks. Traditional cloud computing and edge computing architectures fall short in providing their required flexibility, scalability, and ultra-low latency. Cloud computing centralizes resources in distant data centers, leading to high latency and increased network congestion, while edge computing, though closer to data sources, lacks the agility to dynamically adapt to fluctuating workloads, user mobility, and real-time requirements. In-network computing (INC) offers a transformative solution by integrating computational capabilities directly into the network fabric, enabling dynamic and distributed task execution. This paper explores INC through the lens of information-centric networking (ICN), a revolutionary communication paradigm implementing routing-by-name and in-network caching, and thus emerging as a natural enabler for INC. We review state-of-the-art advancements involving INC and ICN, addressing critical topics such as service naming, executor selection strategies, compute reuse, and security. Furthermore, we discuss key challenges and propose research directions for deploying INC via ICN, thereby outlining a cohesive roadmap for future investigation.

Application of Statistical Models in Air Quality Monitoring

Currently, statistical models are vital instruments for examining and forecasting diverse environmental variables, including air quality analysis. With the development of big data, foundational statistical techniquesincluding regression and time series analysisare adapting to manage larger datasets and complex environmental dynamics. This paper examines the recent applications of statistical models used to assess and forecast air quality in Chinese cities based on existing literature and data research. By focusing on various modeling approaches, such as Poisson regression, grey correlation, and neural networks, it discusses the advantages and drawbacks of each model. Furthermore, it examines the enhancement of classical analytical methodologies inside a big data framework to augment the accuracy of air quality analysis and forecasting. The result shows that although datasets have been more extensive, to increase the accuracy of pollution forecasting, diverse data and more excellent computational capabilities are still needed. Advances in machine learning and optimisation algorithms show promise for overcoming these challenges in the future

All-optical perception based on partially coherent optical neural networks.

In the field of image processing, optical neural networks offer advantages such as high speed, high throughput, and low energy consumption. However, most existing coherent optical neural networks (CONN) rely on coherent light sources to establish transmission models. The use of laser inputs and electro-optic modulation devices at the front end of these neural networks diminishes their computational capability and energy efficiency, thereby limiting their practical applications in object detection tasks. This paper proposes a partially coherent optical neural network (PCONN) transmission model based on mutual intensity modulation. This model does not depend on coherent light source inputs or active electro-optic modulation devices, allowing it to directly compute and infer using natural light after simple filtering, thus achieving full optical perception from light signal acquisition to computation and inference. Simulation results indicate that the model achieves a highest classification accuracy of 96.80% and 86.77% on the MNIST and Fashion-MNIST datasets, respectively. In a binary classification simulation test based on the ISDD segmentation dataset, the model attained an accuracy of 94.69%. It is estimated that this system's computational inference speed for object detection tasks is 100 times faster than that of traditional CONN, with energy efficiency approximately 50 times greater. In summary, our proposed PCONN model addresses the limitations of conventional optical neural networks in coherent light environments and is anticipated to find applications in practical object detection scenarios.

Application of Google Earth Engine to Monitor Greenhouse Gases: A Review

Google Earth Engine (GEE) is a cloud-based platform revolutionizing geospatial analysis by providing access to vast satellite datasets and computational capabilities for monitoring environmental and societal issues. It incorporates machine learning (ML) techniques and algorithms as part of its tools for analyzing and processing large geospatial data. This review explores the diverse applications of GEE in monitoring and mitigating greenhouse gas emissions and uptakes. GEE is a cloud-based platform built on Google’s infrastructure for analyzing and visualizing large-scale geospatial datasets. It offers large datasets for monitoring greenhouse gas (GHG) emissions and understanding their environmental impact. By leveraging GEE’s capabilities, researchers have developed tools and algorithms to analyze remotely sensed data and accurately quantify GHG emissions and uptakes. This review examines progress and trends in GEE applications, focusing on monitoring carbon dioxide (CO2), methane (CH4), and nitrous oxide/nitrogen dioxide (N2O/NO2) emissions. It discusses the integration of GEE with different machine learning methods and the challenges and opportunities in optimizing algorithms and ensuring data interoperability. Furthermore, it highlights GEE’s role in pinpointing emission hotspots, as demonstrated in studies monitoring uptakes. By providing insights into GEE’s capabilities for precise monitoring and mapping of GHGs, this review aims to advance environmental research and decision-making processes in mitigating climate change.

Energy-Efficient Aerial STAR-RIS-Aided Computing Offloading and Content Caching for Wireless Sensor Networks.

Unmanned aerial vehicle (UAV)-based wireless sensor networks (WSNs) hold great promise for supporting ground-based sensors due to the mobility of UAVs and the ease of establishing line-of-sight links. UAV-based WSNs equipped with mobile edge computing (MEC) servers effectively mitigate challenges associated with long-distance transmission and the limited coverage of edge base stations (BSs), emerging as a powerful paradigm for both communication and computing services. Furthermore, incorporating simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RISs) as passive relays significantly enhances the propagation environment and service quality of UAV-based WSNs. However, most existing studies place STAR-RISs in fixed positions, ignoring the flexibility of STAR-RISs. Some other studies equip UAVs with STAR-RISs, and UAVs act as flight carriers, ignoring the computing and caching capabilities of UAVs. To address these limitations, we propose an energy-efficient aerial STAR-RIS-aided computing offloading and content caching framework, where we formulate an energy consumption minimization problem to jointly optimize content caching decisions, computing offloading decisions, UAV hovering positions, and STAR-RIS passive beamforming. Given the non-convex nature of this problem, we decompose it into a content caching decision subproblem, a computing offloading decision subproblem, a hovering position subproblem, and a STAR-RIS resource allocation subproblem. We propose a deep reinforcement learning (DRL)-successive convex approximation (SCA) combined algorithm to iteratively achieve near-optimal solutions with low complexity. The numerical results demonstrate that the proposed framework effectively utilizes resources in UAV-based WSNs and significantly reduces overall system energy consumption.

A Data-Driven Approach to Refine the Partially Stirred Reactor Closure for Turbulent Premixed Flames

AbstractAccurately predicting turbulent combustion processes is fundamental for optimizing efficiency, reducing pollutant emissions, and ensuring operational safety in combustion systems. To this purpose, computational fluid dynamics (CFD) simulations are widely employed. In particular, large eddy simulations (LES) balance prediction accuracy with computational efficiency by resolving only the most energy-containing scales of turbulence and rely on modeling the turbulence-chemistry interactions (TCI) occurring at the smallest scales. Among the existing closures, the partially stirred reactor (PaSR) model incorporates finite-rate chemistry and estimates a cell reacting fraction based on the local Damköhler number to account for the subfilter-scale TCI. Although widely validated in CFD computations, the PaSR model was found limited by the way it computes the cell reacting fraction. To tackle this point, our study proposes a machine learning (ML) enhanced partially stirred reactor model for LES. A fully connected neural network is trained on direct numerical simulation (DNS) data of turbulent premixed jet flames to compute a correction coefficient for the cell reacting fraction. Maintaining the original model shape, this ML-enhanced closure aims at bridging the gap between physics-based models and advanced data-driven techniques. The proposed formulation not only improves the prediction accuracy of quantities of interest such as the heat release rate but also features computational feasibility and generalisation capabilities over a large range of LES grid refinement. This demonstrates the significant potential of ML-aided TCI closures in future applications of combustion engineering.

Characterizing Perception Deep Learning Algorithms and Applications for Vehicular Edge Computing

Vehicular edge computing relies on the computational capabilities of interconnected edge devices to manage incoming requests from vehicles. This offloading process enhances the speed and efficiency of data handling, ultimately boosting the safety, performance, and reliability of connected vehicles. While previous studies have concentrated on processor characteristics, they often overlook the significance of the connecting components. Limited memory and storage resources on edge devices pose challenges, particularly in the context of deep learning, where these limitations can significantly affect performance. The impact of memory contention has not been thoroughly explored, especially regarding perception-based tasks. In our analysis, we identified three distinct behaviors of memory contention, each interacting differently with other resources. Additionally, our investigation of Deep Neural Network (DNN) layers revealed that certain convolutional layers experienced computation time increases exceeding 2849%, while activation layers showed a rise of 1173.34%. Through our characterization efforts, we can model workload behavior on edge devices according to their configuration and the demands of the tasks. This allows us to quantify the effects of memory contention. To our knowledge, this study is the first to characterize the influence of memory on vehicular edge computational workloads, with a strong emphasis on memory dynamics and DNN layers.