The proliferation of computing applications in Edge devices emphasizes the need for efficient and accurate Deep Learning (DL) models, especially in safety applications like Driver Distraction Detection (DDD). However, DL's substantial computational requirements hinder deployment in resourceconstrained environments like vehicles. This paper introduces a differentiable architecture search method for optimal and resource-conscious neural network architecture design. We integrate edge-related constraints in a multi-objective function. We investigate Pareto optimality to explore a diverse set of solutions that cover a spectrum of trade-offs and objectives, rather than a single, narrowly optimized solution. We specifically tailor the model design to target a predetermined computational budget in terms of inference time and model size. The proposed method has been evaluated for DDD using 2 benchmark datasets, namely, SFD and AUCD2, and deployed on a spectrum of devices (workstation, embedded system, and mobile devices). We obtained detection accuracies of 98.17% and 95.80% on SFD and AUCD2, respectively, while significantly reducing model sizes to 0.25 MB and 0.36 MB and inference latency to 3 ms and 4 ms on Nvidia Jetson Xavier NX. Additionally, we achieve almost an order of magnitude fewer parameters (0.06M and 0.09M) compared to state-of-the-art.

Numerical simulations of heat transfer in an electronic control unit (ECU) for an automotive application are developed in Ansys ® Icepak ™. The approach described in this paper sets a framework that allows a comprehensive and integrated thermal analysis of the ECU under varying loading conditions. The approach lends itself to operating point characterization, design optimization, development of reduced order models and subsequent integration with system models. This is desirable since it saves substantial costs associated with sizing of ECUs, power components and development of test setups for iterative design development. Further, the adequacy of PCB (printed circuit board) traces for transient loads is verified in this work which is very valuable for cost savings and product reliability. Experimental validations are performed to support the analysis with constant and varying power losses through the components. The results obtained from the simulations compare very well, within 5% of the results obtained from experiments. The integrated thermal models developed herein provide an accurate estimation of MOSFET junction temperatures which is crucial for power electronics design.

This paper develops a design framework for sensorless  flux-vector control of synchronous machine drives. Observer-based  volts-per-hertz (V/Hz) control is presented as a special case in this framework. A complete linearized model is derived, which allows systematic control design and stability analysis. Using the model, passivity conditions are derived. The control methods are experimentally evaluated using a 6.7-kW synchronous reluctance machine (SyRM) and a 2.2-kW permanent-magnet (PM) machine as examples.

Objective: Deep Brain Stimulation (DBS) is effective for movement disorders, particularly Parkinson's disease (PD). However, a closed-loop DBS system using reinforcement learning (RL) for automatic parameter tuning, offering enhanced energy efficiency and the effect of thalamus restoration, is yet to be developed for clinical and commercial applications. Methods: In this research, we instantiate a basal ganglia-thalamic (BGT) model and design it as an interactive environment suitable for RL models. Four finely tuned RL agents based on different frameworks, namely Soft Actor-Critic (SAC), Twin Delayed Deep Deterministic Policy Gradient (TD3), Proximal Policy Optimization (PPO), and Advantage Actor-Critic (A2C), are established for further comparison. Results: Within the implemented RL architectures, the optimized TD3 demonstrates a significant 67% reduction in average power dissipation when compared to the open-loop system while preserving the normal response of the simulated BGT circuitry. As a result, our method mitigates thalamic error responses under pathological conditions and prevents overstimulation. Significance: In summary, this study introduces a novel approach to implementing an adaptive parameter-tuning closed-loop DBS system. Leveraging the advantages of TD3, our proposed approach holds significant promise for advancing the integration of RL applications into DBS systems, ultimately optimizing therapeutic effects in future clinical trials.

Transformers are widely used in natural language processing and computer vision, and Bidirectional Encoder Representations from Transformers (BERT) is one of the most popular pre-trained transformer models for many applications. This paper studies the dependability and impact of soft errors on BERT implemented with different floating-point formats using two case studies: sentence emotion classification and question & answering. Simulation by error injection is conducted to assess the impact of errors on different parts of the BERT model and different bits of the parameters. The analysis of the results leads to the following findings: 1) in both single and half precision, there is a Critical Bit (CB) on which errors significantly affect the performance of the model; 2) in single precision, errors on the CB may cause overflow in many cases, which leads to a fixed result regardless of the input; 3) in half precision, the errors do not cause overflow but they may still introduce a large accuracy loss. In general, the impact of errors is significantly larger in singleprecision than half-precision parameters. Error propagation analysis is also considered to further study the effects of errors on different types of parameters and reveal the mitigation effects of the activation function and the intrinsic redundancy of BERT.

LoRaWAN is an innovative and prominent communication protocol in the domain of Low Power Wide Area Network (LPWAN), known for its ability to provide long-range communication with low energy consumption. However, the practical implementation of the LoRaWAN protocol, operating at the Medium Access Control (MAC) layer and built upon the LoRa physical (PHY) layer, presents numerous research challenges, including network congestion, interference, optimal resource allocation, collisions, scalability, and security. To mitigate these challenges effectively, the adoption of cutting-edge data-driven technologies such as Deep Learning (DL) and Machine Learning (ML) emerges as a promising approach. Interestingly, very few existing survey or tutorial has addressed the importance of ML or DL-based techniques for LoRaWAN in its current state. This article provides a comprehensive survey of current LoRaWAN challenges and recent solutions, particularly using DL and ML algorithms. The primary objective of this survey is to stimulate further research efforts to enhance the performance of LoRa networks and facilitate their practical deployments. We start by providing a technical background to LoRa alliances, LoRa, and LoRaWAN. Furthermore, we discuss an overview of the most utilized DL and ML algorithms for overcoming LoRaWAN challenges. We also present an interoperable reference architecture for LoRaWAN and validate its effectiveness using a wide range of applications. Additionally, we shed light on several evolving challenges of LoRa and LoRaWAN for the future digital network, along with possible solutions. Finally, we conclude our discussion by briefly summarizing our work.

As human-robot interaction rapidly spreads in numerous fields, the subject of robot acceptance gains increasing importance. Visual similarity to the human body, as occurs for humanoids, is generally not enough to guarantee acceptance in physical interaction, as acceptance directly links to comfort and ergonomics, which are measured in terms of the quality of the robot movement perceived by the human. This paper discusses the connection between comfort and similarity of the robot movement to the human one. By considering the kinematic characterization of human movement, this paper focuses on the time laws of such movements, wherein the end-effector path is prescribed. Based on the sigma-lognormal velocity model, a human-likeness index is defined and used to provide with an a priori characterization of trajectories. Such an index can be used to evaluate the performance of trajectory generation algorithms in producing human-like movements, before they are actually executed. For validation purposes, through physical interaction with a robot, a sample of 38 subjects is asked to compare trajectories and judge about their comfort over three experimental campaigns. The results demonstrate a globally consistent trend between the preference in terms of perceived comfort and the distribution of the suggested human-likeness index.

Simulating believable crowds heavily relies upon the perceived realism and diversity of the agents' behaviors, whilst generating novel crowd simulations highly depends on being able to easily manipulate intuitive behavior parameters. We present P2C (Paths-to-Crowds), a method that parameterizes reference crowd data and enables the simulation of similar behaviors in different environments. This approach enables explainable and fine-grained control of simulations since artists can modify at run-time a small set of intuitive parameters in local regions. We incorporate the integration of four fundamental behaviors: goal-seeking, grouping, interaction with areas of interest, and connectivity-into an existing Reinforcement Learning (RL)-based crowd simulation system, which facilitates the creation of customizable agent behaviors and interactions. To learn a parameter model, we synthesize numerous simulations by sampling the parameter space; we then use these data to learn a model that outputs the underlying parameters. To achieve this, we encode the simulation in a set of 2D maps that encode different measurements such as velocities, occupancy, interpersonal distances, path deviations, etc. The trained model can then be used to infer parameters in localized regions of given crowd data (both real and simulated). This approach enables replication of behavior, transfer to new environments, real-time local control, editing of parameters, and explainability of behaviors (with respect to the fundamental behaviors). We evaluate our model's predictive power on real data and compare it against existing baselines. This along with the accompanying user study reveals P2C's potential in terms of achieving behavior realism and diversity.

The gold standard methods for real-time core temperature (CT) monitoring are invasive and cost-inefficient. The application of Kalman filters for an indirect estimation of CT has been explored in the literature since 2010. This paper presents a comparative study between different state of the art Extended Kalman Filter (EKF) estimation algorithms and a new approach based on a biomimetic human body response pre-emptive mapping concept. In this new method, a mapping model of the physiological response of the heart rate (HR) change to CT increase is pre-applied to the input of the EKF estimation CT procedure in a near real-time manner. The algorithm was trained and tested using two datasets (total participants = 18). The best performing algorithm with this novel pre-emptive mapping achieved in an average Root Mean Squared Error (RMSE) of 0.34°C while the best state of the art EKF model (without pre-emptive mapping) resulted in a RMSE of 0.41°C, leading to a 17% improvement performance of our novel method. Given these favorable outcomes, it is compelling to assess its efficacy on a larger dataset in the near future.

We discuss machine learning and its application to security, microchips and present a testable mathematical hypothesis on how the brain stores memories. In particular we present prototypes for next generation processors and RAM memories.

This paper presents a comprehensive study on rare event estimation in power grids, focusing on state-of-the-art adaptive Monte Carlo algorithms. We compare these methods for the optimal power flow problem in various IEEE benchmark models. Based on the results of our study, we analyze the pros and cons of each adaptive method and investigate their beneficial combinations. Overall, the adaptive effort subset simulation (aE-SuS) method and particle integration methods (PIMs) are promising for high-dimensional reliability analysis. By building on IEEE benchmarks, we provide challenging examples for comparing different emerging methods in static network reliability assessment while revealing improvements for these methods. In particular, we introduce a hybrid approach that combines the strengths of both aE-SuS and annealed PIM. Although this method is not as efficient as aE-SuS, it significantly outperforms crude Monte Carlo and is unbiased. We then employ the aE-SuS method and this hybrid approach for risk assessment of the Texas synthetic power grid, which comprises over 5,000 components, thus showcasing scalability for practical applications.

Recently, head-worn inertial sensors have been proposed to characterize gait. However, only few methods allow for both initial foot contacts detection and stride-by-stride gait speed estimation, and none of them has been validated in real-world settings. In this study, we assessed the performance of a two-step machine learning algorithm to estimate initial foot contacts and speed in realworld conditions with a single inertial sensor attached to the temporal region of the head. A deep learning convolutional network is used to detect gait cycles. Then, gait speed is inferred from the detected gait cycles by a regression model. The models were trained and validated against a multisensor wearable system with data of 15 healthy young adults during both structured and real-world walking trials. The stride detector achieved a high F1-score (> 92%) and a mean absolute error smaller than 40 ms. High correlation between target and predicted speed values (Spearman coefficient > 0.86) and low mean absolute error (< 0.08 m/s) were observed. These findings pave the way to the establishment of gait analysis frameworks based on the integration of inertial sensors with head-worn devices.

The development of automated and connected driving functions is currently a central objective for vehicle manufacturers. Such functions generally are introduced at different levels of automation with limited operational design domains (ODDs), which is gradually extended. However, a concise and practical description of ODDs has not yet been established. This work aims at providing a suitable and mathematically concise description of the operational design domain and relates the new description to the definitions of related terms that are widely used in the research community. This work follows a top-down approach. Engineering applications for ODD descriptions are introduced that go beyond scenario-based test design, like ADAS specification, function delimitation and cooperative, connected mobility. Furthermore the ODD can be seen as an instrument and language for the description of system capabilities, building a fundamental tool for cooperative and collaborative development and operations. A set of requirements on the parameterization of operational design domains is derived and methods for selecting suitable parameters are presented. The application of these methods is demonstrated with real world examples. Finally, a discussion of open issues provides starting points for continuous further research.

This survey provides an overview of the state-of-the-art in runtime adaptive Approximate Computing (AxC) for Deep Learning (DL) algorithms, highlighting the challenges and opportunities in the field. The survey covers a broad spectrum of applications, including medical applications, computer vision, and natural language processing. Various power-constrained platforms, such as System-on-Chips (SoCs), Application Specific Integrated Circuits (ASICs), and Field Programmable Gate Arrays (FPGAs), are explored for their utilization in implementing runtime adaptive AxC. The survey explores various techniques, such as dynamic quantization, adaptive pruning, and low-rank approximations, offering a detailed discussion of their advantages and disadvantages. Specifically, in some surveyed research works, the runtime approximation is achieved through the utilization of machine learning algorithms, with a notable emphasis on Reinforcement Learning (RL). These approaches aim to realize runtime conditions and exploit them appropriately. By providing insights into the advancements and trends in runtime adaptive AxC, this survey serves as a valuable resource for researchers and practitioners interested in this rapidly evolving area of computing. This survey conducts an in-depth investigation into the application, challenges, and scope of runtime adaptive AxC techniques, aiming to mitigate energy consumption while preserving acceptable levels of accuracy in DL models. Our primary focus lies on Convolutional Neural Networks (CNNs), with an emphasis on their application in diverse domains. In striving for comprehensiveness, the survey encompasses selected research works that extend beyond CNNs, including alternative DL models like Recurrent Neural Networks (RNNs). our scope of applications, focuses on CNNs; however, to make a comprehensive survey, we cover some surveyed research works that contain other DL models, such as RNNs. It also highlights the importance of considering specific application requirements and available resources when choosing the appropriate technique.

Artistic Neural Style Transfer (NST) has achieved remarkable success for images. However, this is not the case for dynamic 3D environments, such as computer games, where temporal coherence remains a challenge. Our paper presents an approach that uses the G-buffer information available in a game pipeline to generate robust and temporally consistent in-game artistic stylizations based on a style reference image. We use a synthetic dataset created from open-source computer games and demonstrate that the utilization of depth, normals, and edge information enables the stylization process to be more aware of the geometric and semantic aspects of a game scene. The proposed approach builds on previous work by injecting style transfer in the rendering pipeline, while also utilizing G-buffer information during inference time to improve upon the stability of the stylizations, offering a controllable way to stylize computer games in terms of temporal coherence and content preservation. Qualitative and quantitative evaluations of our in-game stylization network demonstrate significantly higher temporal stability compared to existing style transfer approaches when stylizing 3D computer games.

The emerging Processing-in-Memory (PIM) architecture shows promise in efficiently handling Deep Neural Network (DNN) inference and training by minimizing data movement through analog computing in memory. Unfortunately, the energy efficiency of PIM is still limited since it struggles to efficiently process the massive nonlinear activation functions (AFs) widely used in DNNs. Within the current solutions, the AFs require to be fulfilled in the digital domain with co-processors or lookup-table (LUT) modules. Therefore, a significant amount of data requires a round-trip conversion between the analog and the digital domains at each AF. In this way, the unavoidable analogto-digital and digital-to-analog (AD/DA) conversions dominate the system's power consumption and reduce the overall energy efficiency. To address these issues, we propose a reconfigurable analog module for nonlinear AF computation, named RAM-NAF. RAM-NAF is a pure analog module that utilizes Taylor approximation (TA) to fit the arbitrary AFs, thereby reducing AD/DA conversions. To enhance the accuracy, RAM-NAF adopts a segmentation calculation method (SCM) based on the characteristics of nonlinear AFs. Moreover, the RAM-NAF provides the capability for reconfiguration to support various AFs and can be easily integrated with the existing PIM accelerators. The experiment results show that the proposed RAM-NAF significantly improves the performance of the PIM accelerators and reduces the energy consumption: when performing inference on various PIM accelerators, the energy consumption of AD/DA conversions can be reduced up to 12.31×, while the overall energy efficiency can be increased by 2.34× to 5.20×, and the accuracy loss is below 1%.

This paper introduces a novel approach employing periodic functions for the comprehensive analysis of prime numbers. The method encompasses primality testing, factor counting and listing, prime distribution calculation, and the determination of the Nth prime. The exposition of the technique is presented in a clear and sequential manner, guiding the reader through each step with explicit equations. Graphs are strategically incorporated between crucial stages to facilitate a rapid and intuitive visualization of the rationale and outcomes of each maneuver. The paper concludes with concise reflections and ongoing inquiries into the potential applications and refinements of the proposed method.

This study introduces a hierarchical key assignment scheme (HKAS) based on the closest vector problem in an inner product space (CVP-IPS). The proposed scheme offers a comprehensive solution with scalability, flexibility, cost-effectiveness, and high performance. The key features of the scheme include CVP-IPS based construction, the utilization of two public keys by the scheme, a distinct basis set designated for each class, a direct access scheme to enhance user convenience, and a rigorous mathematical and algorithmic presentation of all processes. This scheme eliminates the need for top-down structures and offers a significant benefit in that the lengths of the basis sets defined for classes are the same and the costs associated with key derivation are the same for all classes, unlike top-down approaches, where the higher class in the hierarchy generally incurs much higher costs. The scheme excels in both vertical and horizontal scalability due to its utilization of the access graph and is formally proven to achieve strong key indistinguishability security (S-KI-security). This research represents a significant advancement in HKAS systems, providing tangible benefits and improved security for a wide range of use cases.