Luca Daniel

<b>Overview</b>This dataset provides multimodal, time-synchronized measurements collected during slow, rhythmic arm reaching. It bridges internal soft-tissue motion and conventional biomechanical measurements by combining B-mode ultrasound imaging of the upper arm with optical motion capture, surface electromyography, and tri-axial accelerometry.Data were collected from 36 healthy adult participants across three expertise levels (expert, intermediate, non-expert). Participants performed unconstrained reaching cycles paced by a visual metronome (1 cycle every 6 s) under two hand-orientation conditions (“give”/palm up; “touch”/palm down). All modalities are time-aligned, and the dataset includes processed signals, derived parameters, event annotations, and metadata to support both biomechanics and motor control analyses, as well as machine learning (ML) work on ultrasound tracking.<b>Key features include:</b>Synchronized motion capture (upper-limb marker trajectories), ultrasound videos, EMG and accelerometry (palm, biceps, triceps), and derived measures (e.g., reach-cycle events, arm kinematics/speed, tremor events/power, EMG amplitude)B-mode ultrasound videos (60 fps) capturing a transverse view of the triceps/brachialis region during reaching.Frame-level ultrasound point trajectories for 11 tracked points (including humerus, muscle boundary, and within muscle points) spanning ~300,000 frames across the dataset.<b>Citation.</b> Please visit the associated dataset descriptor paper for full details and cite it if you use the dataset:<br>Pallarès-López, R., Folgado, D., Magana-Salgado, U. <i>et al.</i> A multimodal biomechanics dataset with synchronized kinematics and internal tissue motions during reaching. <i>Scientific Data</i> <b>13</b>, 709 (2026). https://doi.org/10.1038/s41597-026-07019-3<br><b>Ethics &amp; de-identification. </b>Data are de-identified and shared publicly under written informed consent and MIT COUHES approval (Protocol #2201000537).<b>Contents</b>The dataset is organized as follows:<b>README.txt</b>: high-level documentation and pointers.<b>dataset.csv</b>: one row per participant with demographics/anthropometrics, expertise level, experimental configuration, and paths to the corresponding data files.<b>SHA256SUMS.txt</b>: checksum manifest for integrity verification.<b>data.zip archive</b> (download to obtain all data files in a preserved folder structure), containing:<code>hdf5_files/</code>: 36 (1 per participant) <code>.h5</code> files with synchronized multimodal timeseries, derived signals, event annotations, and metadata.<code>us_videos/</code>: 36 (1 per participant) ultrasound <code>.mp4</code> videos (60 fps).<code>hdf5_structure.txt</code>: description of the HDF5 layout (groups/datasets/attributes).<code>exceptions.txt</code>: participant-specific notes/exceptions (e.g., missing channels or irregularities).<b>HDF5 organization (high-level).</b> Each per-participant HDF5 file follows a consistent structure with (i) a <code>metadata</code> group, (ii) an <code>events</code> group (trial/cycle/tremor annotations), and (iii) a <code>timeseries</code> group containing synchronized sensor streams and derived signals, including subgroups with accelerometry, motion capture, electromyography, ultrasound tracker trajectories, tremor-related measures, arm kinematics/speed, and EMG amplitude. See <code>hdf5_structure.txt</code> for details. For the complete definition of every recorded and derived variable (names, descriptions, units, and array shapes), see the manuscript Supplementary Materials, Section HDF5 File Description.<br><b>How to use</b><b>Start with the dataset.csv</b><b>.</b> Each row corresponds to one participant and includes file paths to the participant’s HDF5 file and ultrasound video.<b>Download and unzip the data.zip archive.</b> The archive preserves the folder structure for <code>hdf5_files/</code> and <code>us_videos/</code>.<b>Read exceptions.txt</b> before running analyses across all participants (it lists known participant-specific exceptions/notes).<b>Load data using the code repository.</b> Tutorials for loading/visualizing the data and reproducing derived measurements are provided in a companion GitHub repository: https://github.com/RogerPallares/reaching-dataset.<b>Potential use cases</b>Supervised training/benchmarking of ultrasound point-tracking models using the provided tracked point trajectories.Development of ultrasound-based metrics for characterizing soft-tissue motion during movement.Biomechanical and motor-control studies linking internal tissue motion to kinematics, tremor measures, muscle activation, and expertise level.Initial insights on expertise from this dataset are available in:Namburi, P., Pallarès-López, R., Folgado, D. <i>et al.</i> Efficient elastic tissue motions indicate general motor skill. <i>Sci Rep</i> <b>15</b>, 36532 (2025). https://doi.org/10.1038/s41598-025-17092-0

<b>Overview</b>This dataset provides multimodal, time-synchronized measurements collected during slow, rhythmic arm reaching. It bridges internal soft-tissue motion and conventional biomechanical measurements by combining B-mode ultrasound imaging of the upper arm with optical motion capture, surface electromyography, and tri-axial accelerometry.Data were collected from 36 healthy adult participants across three expertise levels (expert, intermediate, non-expert). Participants performed unconstrained reaching cycles paced by a visual metronome (1 cycle every 6 s) under two hand-orientation conditions (“give”/palm up; “touch”/palm down). All modalities are time-aligned, and the dataset includes processed signals, derived parameters, event annotations, and metadata to support both biomechanics and motor control analyses, as well as machine learning (ML) work on ultrasound tracking.<b>Key features include:</b>Synchronized motion capture (upper-limb marker trajectories), ultrasound videos, EMG and accelerometry (palm, biceps, triceps), and derived measures (e.g., reach-cycle events, arm kinematics/speed, tremor events/power, EMG amplitude)B-mode ultrasound videos (60 fps) capturing a transverse view of the triceps/brachialis region during reaching.Frame-level ultrasound point trajectories for 11 tracked points (including humerus, muscle boundary, and within muscle points) spanning ~300,000 frames across the dataset.<b>Citation.</b> Please visit the associated dataset descriptor paper for full details and cite it if you use the dataset:<br>Pallarès-López, R., Folgado, D., Magana-Salgado, U. <i>et al.</i> A multimodal biomechanics dataset with synchronized kinematics and internal tissue motions during reaching. <i>Scientific Data</i> <b>13</b>, 709 (2026). https://doi.org/10.1038/s41597-026-07019-3<br><b>Ethics &amp; de-identification. </b>Data are de-identified and shared publicly under written informed consent and MIT COUHES approval (Protocol #2201000537).<b>Contents</b>The dataset is organized as follows:<b>README.txt</b>: high-level documentation and pointers.<b>dataset.csv</b>: one row per participant with demographics/anthropometrics, expertise level, experimental configuration, and paths to the corresponding data files.<b>SHA256SUMS.txt</b>: checksum manifest for integrity verification.<b>data.zip archive</b> (download to obtain all data files in a preserved folder structure), containing:<code>hdf5_files/</code>: 36 (1 per participant) <code>.h5</code> files with synchronized multimodal timeseries, derived signals, event annotations, and metadata.<code>us_videos/</code>: 36 (1 per participant) ultrasound <code>.mp4</code> videos (60 fps).<code>hdf5_structure.txt</code>: description of the HDF5 layout (groups/datasets/attributes).<code>exceptions.txt</code>: participant-specific notes/exceptions (e.g., missing channels or irregularities).<b>HDF5 organization (high-level).</b> Each per-participant HDF5 file follows a consistent structure with (i) a <code>metadata</code> group, (ii) an <code>events</code> group (trial/cycle/tremor annotations), and (iii) a <code>timeseries</code> group containing synchronized sensor streams and derived signals, including subgroups with accelerometry, motion capture, electromyography, ultrasound tracker trajectories, tremor-related measures, arm kinematics/speed, and EMG amplitude. See <code>hdf5_structure.txt</code> for details. For the complete definition of every recorded and derived variable (names, descriptions, units, and array shapes), see the manuscript Supplementary Materials, Section HDF5 File Description.<br><b>How to use</b><b>Start with the dataset.csv</b><b>.</b> Each row corresponds to one participant and includes file paths to the participant’s HDF5 file and ultrasound video.<b>Download and unzip the data.zip archive.</b> The archive preserves the folder structure for <code>hdf5_files/</code> and <code>us_videos/</code>.<b>Read exceptions.txt</b> before running analyses across all participants (it lists known participant-specific exceptions/notes).<b>Load data using the code repository.</b> Tutorials for loading/visualizing the data and reproducing derived measurements are provided in a companion GitHub repository: https://github.com/RogerPallares/reaching-dataset.<b>Potential use cases</b>Supervised training/benchmarking of ultrasound point-tracking models using the provided tracked point trajectories.Development of ultrasound-based metrics for characterizing soft-tissue motion during movement.Biomechanical and motor-control studies linking internal tissue motion to kinematics, tremor measures, muscle activation, and expertise level.Initial insights on expertise from this dataset are available in:Namburi, P., Pallarès-López, R., Folgado, D. <i>et al.</i> Efficient elastic tissue motions indicate general motor skill. <i>Sci Rep</i> <b>15</b>, 36532 (2025). https://doi.org/10.1038/s41598-025-17092-0

We propose a novel projection method that guarantees the conservation of integral quantities in Physics-Informed Neural Networks (PINNs). While the soft constraint that PINNs use to enforce the structure of partial differential equations (PDEs) enables necessary flexibility during training, it also permits the discovered solution to violate physical laws. To address this, we introduce a projection method that guarantees the conservation of the linear and quadratic integrals, both separately and jointly. We derived the projection formulae by solving constrained non-linear optimization problems and found that our PINN modified with the projection, which we call PINN-Proj, reduced the error in the conservation of these quantities by three to four orders of magnitude compared to the soft constraint and marginally reduced the PDE solution error. We also found evidence that the projection improved convergence through improving the conditioning of the loss landscape. Our method holds promise as a general framework to guarantee the conservation of any integral quantity in a PINN if a tractable solution exists.

Insights into the general nature of motor skill could fundamentally change how we develop movement abilities, with implications for musculoskeletal well-being and injury. Here, we sought to identify indicators of general motor skill-those shared by experts across disciplines (e.g., squash, ballet, volleyball) during non-specialized movements (e.g., reaching for water). Identifying such general indicators of motor skill has remained elusive. Using ultrasound imaging with deep learning and optical flow analysis, we tracked elastic tissues (muscles and associated connective tissues) during a simple reaching task performed similarly by world-class athletes and regional-level athletes drawn from diverse disciplines, as well as untrained non-experts. We analyzed two types of inefficient tissue motions that do not contribute to the net work done by the muscles to actuate joints. These are transverse muscle movements orthogonal to the muscle fiber direction and physiological tremors. We discovered that world-class experts minimize both of these inefficient motions compared to regional-level athletes and non-experts. While regional-level athletes surprisingly showed similar inefficiencies to non-experts, they used elastic tissues more effectively, achieving equivalent arm movements with smaller actuation-related tissue motions. We establish elastic tissue motion as a key indicator of general motor skill, expanding our understanding of elastic mechanisms and their role in general aspects of motor skill.

Electromagnetic (EM) fields are widely used in several applications where maximizing penetration into the target object is desirable, such as biomedical imaging and sensing. This improvement is often achieved through a matching layer (ML), which can be either a dielectric material or dielectrics patterned with metallic structures. A metasurface ML (MML) represents an improvement over traditional dielectric MLs (DMLs) since it provides better impedance matching, leading to increased field strength within the target and reduced ML thickness. However, designing a MML for near-field EM sources has traditionally relied on time-consuming full-wave simulations. This article introduces a novel analytical approach to designing an MML for arbitrary near-field EM sources, providing a more efficient and effective solution. Three cases are analyzed to assess the reliability of the proposed approach: a short dipole, a small loop, and a patch antenna. The performance of the MML, obtained by an analytical procedure, is preliminarily assessed with full-wave simulations and then experimentally validated. The optimized MML is fabricated using 3D and 2D additive manufacturing (AM) techniques. A comparative analysis across all the considered scenarios confirms the robustness and reliability of the proposed MML design strategy.

Accurately predicting the energy consumption of Battery Electric Vehicles (BEVs) is essential for addressing range anxiety, optimizing route planning, and governing infrastructure investments in a rapidly electrifying transportation sector. This paper presents a generalized, flexible, and probably the most comprehensive modeling framework designed to estimate BEV energy consumption under different driving conditions, vehicle configurations, and environmental influences. The model is structured in mechanical, electrical, and auxiliary sub-models. The model incorporates detailed input parameters, such as aerodynamic coefficients, transmission and motor characteristics, regenerative braking constraints, battery capacity, climate control demands, and ambient conditions. Validation results demonstrate a strong alignment between measured and predicted power profiles, with a high coefficient of determination, low RMSE, and low MAE confirming the model’s reliability and adaptability. The introduced framework can be extended to various BEV segments, driving cycles, and environmental conditions, providing valuable information for vehicle manufacturers, fleet managers, and policymakers aiming to improve EV performance, route efficiency, and charging infrastructure deployment.

In pursuit of petahertz electronics, we seek to develop light-wave electronic circuits which are orders of mag-nitude faster than conventional electronics and work to realize signal processing at unprecedented bandwidths [1]. Optically driven nanoantennas are a promising candidate for petahertz electronics, with many advantages such as sub-cycle attosecond charge transport, polarization sensitivity, and low optical field requirements due to their geometrical and resonant field enhancement [2]–[4]. However, the development of petahertz logic-gates [5] and memory circuits is hindered by the computational cost of full electromagnetic and particle tracking simulations, which become unwieldy when scaling from a single antenna to a system of multiple interconnected ones. To overcome this challenge, we developed a compact circuit model in LTspice which goes beyond describing the electromagnetic response as was done in prior works [6]. Our model also describes charge transport in a nanoantenna, paving the way for effective modeling of functional petahertz circuit elements.

The state-of-the-art (SOTA) deep learning based time series models are inspired by convolutional neural networks (CNN), recurrent neural networks (RNN) or transformers which are successful architectures for domains like vision, text, etc. However, the gold standard architecture for time series modeling is not yet established. In this paper, we propose a new neural network structure that can be used as a strong baseline for time series problems, leveraging dilated kernels with fully convolutional networks (FCNs). The proposed model, called the dilated multi-kernel fully convolutional network (DM-FCN), is a composite model that leverages a vast receptive field and is designed to capture the long-distance interaction in multivariate time series data. We evaluate the performance of the DM-FCN model on a variety of time series benchmarks. Our results show that the baseline DM-FCN model outperforms state-of-the-art models on many of the benchmarks by a large margin. By integrating statistical insights, we also evaluated different variations of DM-FCN and deliberated on model selections across diverse time series data.

Abstract Insights into the general nature of motor skill could fundamentally change how we develop movement abilities, with implications for musculoskeletal well-being and injury. Here, we sought to identify indicators of general motor skill–those shared by experts across disciplines (e.g., squash, ballet, volleyball) during non-specialized movements (e.g., reaching for water). Identifying such general indicators of motor skill has remained elusive. Using ultrasound imaging with deep learning and optical flow analysis, we tracked elastic tissues (muscles and associated connective tissues) during a simple reaching task performed similarly by world-class athletes and regional-level athletes drawn from diverse disciplines, as well as untrained non-experts. We analyzed previously unexamined inefficiencies: transverse elastic tissue motions orthogonal to muscle fiber direction and physiological tremors, which are oscillations that do not contribute to the net work done by muscles. We discovered that world-class experts minimize both these inefficient motions compared to regional level athletes and non-experts. While regional- level athletes surprisingly showed similar inefficiencies to non-experts, they used elastic tissues more effectively, achieving equivalent movements with smaller actuation-related tissue motions. We establish elastic tissue motion as a key indicator of general motor skill, expanding our understanding of elastic mechanisms and their role in general aspects of motor skill.

This study presents a practical approach to quantify the economic benefits of dynamic price-based charging strategies for electric bus operators with overnight depot charging. The replication of the methodology allows for the precise estimation and charging schedule for the specific schedule. However, this work is also designed to target decision makers, allowing them to access the investment in this technology and its integration using the presented results. This paper defines a Mixed-Integer Linear Programming framework to optimize electric bus charging schedules using day-ahead electricity prices, incorporating real-world operational constraints and infrastructure limitations, and presents the potential savings achievable with the implementation of the proposed framework. The methodology includes a benchmark that schedules the charging as soon as possible, thereby replicating uncontrolled charging. Validated with actual operational data from a medium-sized Italian transit company, the model demonstrates substantial cost reductions by strategically shifting charging times to exploit lower-priced energy periods, especially aligning overnight charging with reduced electricity tariffs. The case study with real data from Italy shows a potential reduction of up to 1500 € (9% of the bill) for an 11-bus fleet. Scenario-based analyses highlight potential monthly savings with reductions in energy costs ranging between 7.8% and 15.1%, depending on operational contexts. Sensitivity analyses confirm that the operational variability further enhances cost-saving opportunities, while constrained grid power significantly limits these benefits, showing the impact for an adequately sized charging infrastructure. Also, a table is provided to easily extend the results to other European countries.

Presents corrections to the paper, Electron Emission Regimes of Planar Nano Vacuum Emitters.

Physics-informed neural networks (PINNs) incorporate physical laws into their training to efficiently solve partial differential equations (PDEs) with minimal data. However, PINNs fail to guarantee adherence to conservation laws, which are also important to consider in modeling physical systems. To address this, we proposed PINN-Proj, a PINN-based model that uses a novel projection method to enforce conservation laws. We found that PINN-Proj substantially outperformed PINN in conserving momentum and lowered prediction error by three to four orders of magnitude from the best benchmark tested. PINN-Proj also performed marginally better in the separate task of state prediction on three PDE datasets.

We introduce physics-informed Fourier networks (PIFNs) for Electrical Properties (EP) Tomography (EPT). Our novel deep learning-based method is capable of learning EPs globally from noisy magnetic resonance (MR) measurements, i.e, the magnitude of the magnetic transmit field and the transceive phase. Our proposed method also provides noise-free transmit field reconstructions. Two separate Fourier neural networks are used to efficiently estimate the transmit field and EPs at any location. We show that PIFN EPT accurately infers the EPs distribution of an inhomogeneous phantom from noisy simulated measurements.

We propose a novel approach to electrical property (EP) estimation that also estimates the incident fields generated by a coil, by solving a basis pursuit problem. We also propose a novel Global Maxwell Tomography (GMT) formulation that uses the MR signal instead of $$$B_1^+$$$. We tested our approach experimentally by reconstructing the average EP of a homogeneous cylinder. We obtained &lt; 5% estimation error using only $$$B_1^+$$$ data obtained from MR Fingerprinting. We then estimated the receive sensitivity maps $$$B_1^-$$$ by using the new signal-based GMT.

We present the first phase retrieval algorithm guaranteed to solve the multidimensional phase retrieval problem in polynomial arithmetic complexity without prior information. The method successfully terminates in O(N log(N)) operations for Fourier measurements with cardinality N. The algorithm is guaranteed to succeed for a large class of objects, which we term "Schwarz objects". We further present an easy-to-calculate and well-conditioned diagonal operator that transforms any feasible phase-retrieval instance into one that is solved by our method. We derive our method by combining techniques from classical complex analysis, algebraic topology, and modern numerical analysis. Concretely, we pose the phase retrieval problem as a multiplicative Cousin problem, construct an approximate solution using a modified integral used for the Schwarz problem, and refine the approximate solution to an exact solution via standard optimization methods. We present numerical experimentation demonstrating our algorithm's performance and its superiority to existing method. Finally, we demonstrate that our method is robust against Gaussian noise.

Accurate estimation of rare failure occurrence probability is crucial for ensuring the proper and reliable functioning of integrated circuits (ICs). Conventional Monte Carlo methods are inefficient, demanding an exorbitant number of samples to achieve reliable estimates. Inspired by the exact sampling capabilities of normalizing flows, we revisit this problem and propose normalizing flow assisted importance sampling, termed NOFIS. NOFIS first learns a sequence of proposal distributions associated with predefined nested subset events by minimizing KL divergence losses. Next, it estimates the rare event probability by utilizing importance sampling in conjunction with the last proposal. The efficacy of our NOFIS method is substantiated through comprehensive qualitative visualizations, affirming the optimality of the learned proposal distribution, as well as 10 quantitative experiments, which highlight NOFIS's superior accuracy over baseline approaches.

Neural networks are powerful tools in various applications, and quantifying their uncertainty is crucial for reliable decision-making. In the deep learning field, the uncertainties are usually categorized into aleatoric (data) and epistemic (model) uncertainty. In this paper, we point out that the existing popular variance attenuation method highly overestimates aleatoric uncertainty. To address this issue, we proposed a new estimation method by actively de-noising the observed data. By conducting a broad range of experiments, we demonstrate that our proposed approach provides a much closer approximation to the actual data uncertainty than the standard method.

Across-task binding is defined as the stimulus/response of one task being linked to the response of another task. The purpose of the present experiment was to determine across-task binding in a continuous movement sequence task with an auditory task of high and low pitch tones and the development of a movement sequence representation. According to the two systems theory of sequence learning, we expected that the developed representation in the across-task binding context relies on the multi-dimensional system rather than on the unidimensional system which is restricted to a set of modules where each module processed information along one task/dimension. An inter-manual transfer design was used to disentangle the sequence representations. The mirror transfer test required the same pattern of muscle activation and joint angles (motor coordinates) in the contralateral limb as experienced during the acquisition phase, while in the non-mirror transfer test, the visual-spatial locations (spatial coordinates) of the target waveform were reinstated. The main finding was that consistently combining visual-spatial positions in a sequence and auditory dimensions such as the tone pitch does not rely on a multidimensional system as predicted by the two-systems theory.

.This work aims to accelerate the convergence of proximal gradient methods used to solve regularized linear inverse problems. This is achieved by designing a polynomial-based preconditioner that targets the eigenvalue spectrum of the normal operator derived from the linear operator. The preconditioner does not assume any explicit structure on the linear function and thus can be deployed in diverse applications of interest. The efficacy of the preconditioner is validated on three different Magnetic Resonance Imaging applications, where it is seen to achieve faster iterative convergence (around \(\mbox 2\!-\!3\times\) faster, depending on the application of interest) while achieving similar reconstruction quality.Keywordsregularized linear inverse problemspolynomial preconditionerproximal gradient descentMSC codes90C0690C9090C2592C55

Electric vehicle (EV) adoption has been increasing rapidly, posing new challenges for integrating EV charging infrastructure with the existing electrical grid. Uncoordinated charging of EVs can cause transformers to overload, leading to instability and unreliability in the grid. This article introduces two smart charging coordinators for EV charging pools designed to manage EV charging while considering transformer power limits. The first strategy aims to minimize operational costs, while the second maximizes the charger flexibility. Both coordinators account for uncertainties in EV arrival time and state of charge, as well as inflexible demands on transformers. The strategies are evaluated and compared using grid-aware and grid-unaware methods regarding transformer power limits. Real-world datasets are utilized to assess the performance of the proposed strategies through simulation studies across three scenarios: single charging station behavior, average parking lot occupancy, and worst-case occupancy scenarios. Comparative analysis against uncoordinated and coordinated strategies from the literature reveals that the flexibility maximization strategy provides the most uniform response, effectively mitigating transformer overload events by optimizing charging power and scheduling flexibility. The study underscores the importance of accurate, innovative charging strategies for seamless EV integration and emphasizes the necessity of coordinated charging pools for reliable EV charging operations.

We developed a circuit model for petahertz electronic nanoantenna networks. This approach enables fast and scalable simulations of the the attosecond to femtosecond charge dynamics within the nanoantenna networks. We use the model to explore designs for a memory cell and shift register.

We propose Physics-Informed Fourier Networks for Electrical Properties (EP) Tomography (PIFON-EPT), a novel deep learning-based method for EP reconstruction using noisy and/or incomplete magnetic resonance (MR) measurements. Our approach leverages the Helmholtz equation to constrain two networks, responsible for the denoising and completion of the transmit fields, and the estimation of the object's EP, respectively. We embed a random Fourier features mapping into our networks to enable efficient learning of high-frequency details encoded in the transmit fields. We demonstrated the efficacy of PIFON-EPT through several simulated experiments at 3 and 7 tesla(T) MR imaging, and showed that our method can reconstruct physically consistent EP and transmit fields. Specifically, when only 20% of the noisy measured fields were used as inputs, PIFON-EPT reconstructed the EP of a phantom with ≤ 5% error, and denoised and completed the measurements with ≤ 1% error. Additionally, we adapted PIFON-EPT to solve the generalized Helmholtz equation that accounts for gradients of EP between inhomogeneities. This yielded improved results at interfaces between different materials without explicit knowledge of boundary conditions. PIFON-EPT is the first method that can simultaneously reconstruct EP and transmit fields from incomplete noisy MR measurements, providing new opportunities for EPT research.

Neural networks are powerful tools in various applications, and quantifying their uncertainty is crucial for reliable decision-making. In the deep learning field, the uncertainties are usually categorized into aleatoric (data) and epistemic (model) uncertainty. In this paper, we point out that the existing popular variance attenuation method highly overestimates aleatoric uncertainty. To address this issue, we propose a new estimation method by actively de-noising the observed data. By conducting a broad range of experiments, we demonstrate that our proposed approach provides a much closer approximation to the actual data uncertainty than the standard method.

A rare event is defined by a low probability of occurrence. Accurate estimation of such small probabilities is of utmost importance across diverse domains. Conventional Monte Carlo methods are inefficient, demanding an exorbitant number of samples to achieve reliable estimates. Inspired by the exact sampling capabilities of normalizing flows, we revisit this challenge and propose normalizing flow assisted importance sampling, termed NOFIS. NOFIS first learns a sequence of proposal distributions associated with predefined nested subset events by minimizing KL divergence losses. Next, it estimates the rare event probability by utilizing importance sampling in conjunction with the last proposal. The efficacy of our NOFIS method is substantiated through comprehensive qualitative visualizations, affirming the optimality of the learned proposal distribution, as well as a series of quantitative experiments encompassing $10$ distinct test cases, which highlight NOFIS's superiority over baseline approaches.

The balancing of three-phase node voltages in modern power distribution grids can be significantly deteriorated by the penetration of single-phase PV renewable sources. For a given grid topology and prescribed loads, voltage unbalance critically depends on the nodes where power is injected. Its amount can vary substantially at different observations Buses in the grid. In this paper, we present a methodology that can inform network operators about the critical Buses in the grid and critical injection scenarios. The method is based on a numerically efficient but accurate probabilistic load flow that can handle the case of many PV sources and provides detailed information on the probability distribution of voltage unbalance. The proposed methodology relies on the complex-domain modeling of voltage unbalance sensitivity and on accelerating Monte Carlo simulations via parameter space partitioning.

Electromagnetic devices often require the maximization of the transmitted field, especially for biomedical applications. The employment of a matching layer can reduce the reflections at the air/skin interface, but it is often realized without design guidelines and it may result bulky. On the other hand, introducing a metasurface can improve user’s comfort. A rigorous procedure to find the parameters of the matching layer, also comprising the metasurface, can help in its design. Here, we extend the analytical optimization procedure of our previous works by considering a real patch antenna instead of a short dipole. An experimental setup for retrieving the electric field inside a tissue-mimicking phantom is also presented. Three cases are considered: no matching layer, dielectric-only matching layer and metasurface matching layer. The preliminary results show a good agreements with the theoretical ones.

We developed a simple lumped-element circuit model for a resonant vacuum nanoantenna driven by beyond 100 THz optical excitations. The circuit model was implemented in SPICE for fast, easy, and accessible simulation of integrated optical-frequency-nanoantenna circuits. We also designed and simulated an ultrafast memory cell with read and write functions. Our work demonstrates that vacuum nanoantennas can enable logic and memory operations at beyond 100 THz frequencies.

Electric Vehicles (EV) are becoming important players in the energy transition. These vehicles can provide flexible capacity to the electrical grid for maintaining the energy balance. Then, important challenges appear when defining EV flexibility. EV flexibility is usually defined before providing the energy service, ignoring the real-time power dispatch. However, in this paper, we define the evolution of the upward and downward EV flexibility capacities, considering the real-time power dispatch. The flexibility is computed with the energy that can be shifted into the service time. The flexibility is evaluated in three different charging strategies that are designed to avoid the rebound effect. The strategies are applied to EV chargers in the residential, workplace, and public sectors. Three real databases of EV charging events are assessed, one per sector. Simulation results present that the residential sector has more flexibility while the public sector has lower flexibility. This research allows an understanding of the real flexibility each sector can provide to the electrical grid operators taking into account the hours and capacities information.

This paper aims to present a general-purpose Surrogate Model for the probabilistic analysis of power distribution grids with a large number of input parameters. The distinctive feature of the novel technique is the employment of the partial derivatives of output variables versus input parameters to tame the “curse of dimensionality” problem exhibited by prior surrogate model calculation techniques. The second important feature of the proposed Surrogate Model method is that it does not require any a priori assumption about the nature or statistical distribution of the input parameters. In fact, it can be applied whenever design parameters are deterministic variables as well as when they are uncertain and represented by continuous and/or discrete random variables. Relevant applications presented in the paper refer to the probabilistic analysis of the distribution grid in the presence of a large number of photovoltaic sources and electric vehicle charging stations.

We propose Physics-Informed Fourier Networks for Electrical Properties (EP) Tomography (PIFON-EPT), a novel deep learning-based method for EP reconstruction using noisy and/or incomplete magnetic resonance (MR) measurements. Our approach leverages the Helmholtz equation to constrain two networks, responsible for the denoising and completion of the transmit fields, and the estimation of the object's EP, respectively. We embed a random Fourier features mapping into our networks to enable efficient learning of high-frequency details encoded in the transmit fields. We demonstrated the efficacy of PIFON-EPT through several simulated experiments at 3 and 7 tesla (T) MR imaging, and showed that our method can reconstruct physically consistent EP and transmit fields. Specifically, when only $20\%$ of the noisy measured fields were used as inputs, PIFON-EPT reconstructed the EP of a phantom with $\leq 5\%$ error, and denoised and completed the measurements with $\leq 1\%$ error. Additionally, we adapted PIFON-EPT to solve the generalized Helmholtz equation that accounts for gradients of EP between inhomogeneities. This yielded improved results at interfaces between different materials without explicit knowledge of boundary conditions. PIFON-EPT is the first method that can simultaneously reconstruct EP and transmit fields from incomplete noisy MR measurements, providing new opportunities for EPT research.

Deep neural networks (DNN) have shown great capacity of modeling a dynamical system; nevertheless, they usually do not obey physics constraints such as conservation laws. This paper proposes a new learning framework named ConCerNet to improve the trustworthiness of the DNN based dynamics modeling to endow the invariant properties. ConCerNet consists of two steps: (i) a contrastive learning method to automatically capture the system invariants (i.e. conservation properties) along the trajectory observations; (ii) a neural projection layer to guarantee that the learned dynamics models preserve the learned invariants. We theoretically prove the functional relationship between the learned latent representation and the unknown system invariant function. Experiments show that our method consistently outperforms the baseline neural networks in both coordinate error and conservation metrics by a large margin. With neural network based parameterization and no dependence on prior knowledge, our method can be extended to complex and large-scale dynamics by leveraging an autoencoder.

Interpreting machine learning models is challenging but crucial for ensuring the safety of deep networks in autonomous driving systems. Due to the prevalence of deep learning based perception models in autonomous vehicles, accurately interpreting their predictions is crucial. While a variety of such methods have been proposed, most are shown to lack robustness. Yet, little has been done to provide certificates for interpretability robustness. Taking a step in this direction, we present CORGI, short for Certifiably prOvable Robustness Guarantees for Interpretability mapping. CORGI is an algorithm that takes in an input image and gives a certifiable lower bound for the robustness of the top k pixels of its CAM interpretability map. We show the effectiveness of CORGI via a case study on traffic sign data, certifying lower bounds on the minimum adversarial perturbation not far from (4-5x) state-of-the-art attack methods.