Direct Mapping Research Articles

Reconfigurable Acceleration of Neural Networks: A Comprehensive Study of FPGA-based Systems

This paper explores the potential of Field-Programmable Gate Arrays (FPGAs) for accelerating both neural network inference and training. We present a comprehensive analysis of FPGA-based systems, encompassing architecture design, hardware implementation strategies, and performance evaluation. Our study highlights the advantages of FPGAs over traditional CPUs and GPUs for neural network workloads, including their inherent parallelism, reconfigurability, and ability to tailor hardware to specific network needs. We delve into various hardware implementation strategies, from direct mapping to dataflow architectures and specialized hardware blocks, examining their impact on performance. Furthermore, we benchmark FPGA-based systems against traditional platforms, evaluating inference speed, energy efficiency, and memory bandwidth. Finally, we explore emerging trends in FPGA-based neural network acceleration, such as specialized architectures, efficient memory management techniques, and hybrid CPU-FPGA systems. Our analysis underscores the significant potential of FPGAs for accelerating deep learning applications, particularly those requiring high performance, low latency, and energy efficiency.

Spatiotemporal protein interactome profiling through condensation-enhanced photocrosslinking.

Resolving protein-protein interactions (PPIs) inside biomolecular condensates is crucial for elucidating their functions and regulation mechanisms. The transient nature of condensates and the multiple localizations of clients, however, have rendered it challenging to determine compartment-specific PPIs. Here we developed a condensation-enhanced, spatially directed, metabolic incorporation-assisted photocrosslinking strategy-termed DenseMAP-for spatiotemporally resolved dissection of the direct protein interactome within condensates. By leveraging our condensation-enhanced photocrosslinker and the spatially directed biotin tagging, DenseMAP enabled stress-granule-specific interactome mapping of the N6-methyladenosine readers YTHDF1 and YTHDF2, and uncovered the functional role of phosphorylation on the SARS-CoV-2 nucleocapsid protein in regulating virus replication. Further applying DenseMAP for direct interactome mapping of the subcompartmental scaffold protein NPM1 deciphered nucleolar granular component proteome, and unveiled the critical role of SUMOylation in controlling nucleolar proteome homeostasis. DenseMAP provides a platform technology for analysing functional PPI networks within subcellular and subcompartmental condensates under diverse physiological and/or pathological settings.

Efficient aerodynamic shape optimization using transfer learning based multi-fidelity deep neural network

Computational efficiency and precision pose a classic contradiction in aerodynamic shape optimization. To address this challenge, this study introduces an effective optimization framework based on multi-fidelity fully connected neural network (MFFCN). The framework utilizes transfer learning (TL) to train a multi-fidelity surrogate model that establishes direct mappings between geometric configuration parameters and aerodynamic performance by adaptively capturing linear or nonlinear relationships concealed between high-fidelity (HF) and low-fidelity (LF) information. The HF and LF data are derived from fine and coarse grids, respectively, evaluated using the same computational fluid dynamics (CFD) model. The MFFCN-TL framework is applied to optimize the National Advisory Committee for Aeronautics 0012 (NACA0012) airfoil (12 design variables) and the Office National d'Études et de Recherches Aérospatiales M6 (ONERA M6) wing (50 design variables). Simulation results demonstrate that the NACA0012 airfoil achieves a 69.47% enhancement in lift–drag ratio in 1.069 s, compared to a 12.76% gain over 24.8 h in single-fidelity CFD-based optimization. The ONERA M6 wing achieves a 24.66% reduction in drag coefficient in 694 ms compared to 18.37% over 237.3 h in the CFD model. Statistical results show that the MFFCN-TL framework can reduce optimization cost by more than 90% compared to the single-fidelity CFD-based model. These findings suggest that the MFFCN-TL framework significantly enhances optimization efficiency and provides superior feasible solutions over single-fidelity methods.

First test beam of the DMAPS-based proton tracker at the pMBRT facility at the Curie Institute

Objective.Proton radiotherapy's efficacy relies on an accurate relative stopping power (RSP) map of the patient to optimise the treatment plan and minimize uncertainties. Currently, a conversion of a Hounsfield Units map obtained by a common x-ray computed tomography (CT) is used to compute the RSP. This conversion is one of the main limiting factors for proton radiotherapy. To bypass this conversion a direct RSP map could be obtained by performing a proton CT (pCT). The focal point of this article is to present a proof of concept of the potential of fast pixel technologies for proton tracking at clinical facilities.Approach.A two-layer tracker based on the TJ-Monopix1, a depleted monolithic active pixel sensor (DMAPS) chip initially designed for the ATLAS, was tested at the proton minibeam radiotherapy beamline at the Curie Institute. The chips were subjected to 100 MeV protons passing through the single slit collimator (0.4×20mm2) with fluxes up to1.3×107p s-1 cm-2. The performance of the proton tracker was evaluated with GEANT4 simulations.Main results.The beam profile and dispersion in air were characterized by an opening of 0.031 mm cm-1, and aσx=0.172mm at the position of the slit. The results of the proton tracking show how the TJ-Monopix1 chip can effectively track protons in a clinical environment, achieving a tracking purity close to 70%, and a position resolution below 0.5 mm; confirming the chip's ability to handle high proton fluxes with a competitive performance.Significance.This work suggests that DMAPS technologies can be a cost-effective alternative for proton imaging. Additionally, the study identifies areas where further optimization of chip design is required to fully leverage these technologies for clinical ion imaging applications.

Delineating cysteine-reactive compound modulation of cellular proteostasis processes.

Covalent modulators and covalent degrader molecules have emerged as drug modalities with tremendous therapeutic potential. Toward realizing this potential, mass spectrometry-based chemoproteomic screens have generated proteome-wide maps of potential druggable cysteine residues. However, beyond these direct cysteine-target maps, the full scope of direct and indirect activities of these molecules on cellular processes and how such activities contribute to reported modes of action, such as degrader activity, remains to be fully understood. Using chemoproteomics, we identified a cysteine-reactive small molecule degrader of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) nonstructural protein 14 (nsp14), which effects degradation through direct modification of cysteines in both nsp14 and in host protein disulfide isomerases. This degrader activity was further potentiated by generalized electrophile-induced global protein ubiquitylation, proteasome activation and widespread aggregation and depletion of host proteins, including the formation of stress granules. Collectively, we delineate the wide-ranging impacts of cysteine-reactive electrophilic compounds on cellular proteostasis processes.

Digital retailing practices for triggering physical retailers’ bounce-back and bounce-forward performance against a great shock: evidence from the COVID-19 pandemic

This paper explores the mapping from specific digitalization practices to specific resilient performance against a great shock. Based on an adapted structure–conduct–performance framework, this paper hypothesizes by theoretically analyzing how the pre-shock establishment of digital retailing practices could trigger physical retailers’ bounce-back and bounce-forward performance against the COVID-19 crisis. Using a difference-in-differences strategy, the hypotheses are examined with a sample including 549 observations of 50 Chinese listed retailers from the first quarter of 2018 to the third quarter of 2020. The empirical results mainly indicate the following binary findings. First, regardless of the differences in triggering bounce-back performance, different digital retailing practices are found to be effective in triggering physical retailers’ bounce-forward performance against the COVID-19 crisis. This somewhat addresses the concern about the temporality of digitalization-enabled resilience by revealing the generality across digital retailing practices in the sense of triggering resilient performance. Second, it is shown that physical retailers’ bounce-back performance at a specific stage of the COVID-19 crisis could only be triggered by digital retailing practices that coincidentally apply to the shock-induced market structure changes at the stage. The results emphasize each digitalization practice’s individuality in triggering resilient performance. This justifies the non-negligibility of the direct mapping from specific digitalization practices to specific resilient performance in digitalization-enabled resilience evaluation.

Longitudinal In Vivo Imaging of Retinal Cells with Tailored Adeno-Associated Virus Serotypes via Confocal Scanning Laser Ophthalmoscopy.

The dynamic nature of retinal cellular processes necessitates advancements in gene delivery and live monitoring techniques to enhance the understanding and treatment of ocular diseases. This study introduces an optimized adeno-associated virus (AAV) approach, utilizing specific serotypes and promoters to achieve optimal transfection efficiency in targeted retinal cells, including retinal ganglion cells (RGCs) and Müller glia. Leveraging the precision of confocal scanning laser ophthalmoscopy (CSLO), this work presents a non-invasive method for in vivo imaging that captures the longitudinal expression of AAV-mediated green fluorescent protein (GFP). This approach eliminates the need for terminal procedures, preserving the continuity of observation and the well-being of the subject. Furthermore, the GFP signal can be traced in AAV-infected RGCs along the visual pathway to the superior colliculus (SC) and lateral geniculate nucleus (LGN), enabling the potential for direct visual pathway mapping. These findings provide a detailed protocol and demonstrate the application of this powerful tool for real-time studies of retinal cell behavior, disease pathogenesis, and the efficacy of gene therapy interventions, offering valuable insights into the living retina and its connections.

Catch the Cyber Thief: A Multi-Dimensional Asymmetric Network Attack–Defense Game

This paper presents a novel multi-dimensional asymmetric game model for network attack–defense decision-making, called “Catch the Cyber Thief”. The model is built upon the concept of partially observable stochastic games (POSG) and is designed to systematically incorporate multi-dimensional asymmetry into network attack–defense problems. The attack agent is called a “thief” who wants to control a key host by exploring the unfamiliar network environment, and the defense agent is called a “police” who needs to catch the opponent before its goal is accomplished. The results indicate that the asymmetry of network attack and defense is not only influenced by attack and defense strategies but also by spatio-temporal factors such as the attacker’s initial position, network topology, and defense lag time. In addition, we have found that there may exist the “attack rhythm,” which makes “how to maintain a good attack rhythm” and “how to generate a robust defense strategy against different attackers” worth exploring. Compared with existing attack–defense game models, our game model can better generate a direct mapping relationship with real elements, enabling us to understand network attack and defense interactions better, recognize security risks, and design defense strategies that can directly serve real-world decision-making.

Direct Mapping of Polyclonal Epitopes in Serum by HDX-MS.

Elucidating the interactions that drive antigen recognition is central to understanding antibody-mediated protection and is vital for the rational design of immunogens. Often, structural knowledge of epitopes targeted by antibodies is derived from isolated studies of monoclonal antibodies, for which numerous structural techniques exist. In contrast, there are very few approaches capable of mapping the full scope of antigen surfaces targeted by polyclonal sera through the course of a natural antibody response. Here, we develop an approach using immobilized antigen coupled to hydrogen/deuterium exchange with mass spectrometry (HDX-MS) to probe epitope targeting in the context of the fully native serum environment. Using the well-characterized Staphylococcal enterotoxin B (SEB) as a model system, we show that complex combinations of epitopes can be detected and subtle differences across different antisera can be discerned. This work reveals new insight into how neutralizing antibodies and antisera target SEB and, more importantly, establishes a novel method for directly mapping the epitope landscape of polyclonal sera.

Shock buffet onset prediction with flow feature-informed neural network

Transonic shock buffet is a significant self-excited shock oscillations and aerodynamic instability phenomenon induced by shock-boundary layer interaction, which limits the flight envelope and even causes flight accidents. The aviation industry has a significant interest in accurately predicting the shock buffet onset boundary, defined by a specific combination of Mach number and angle of attack. While the current methods of steady and unsteady numerical simulation suffer from a contradiction of efficiency and accuracy. In the current paper, a flow feature-informed neural network (FINN) model is constructed to predict the buffet onset boundary over airfoils. Typical features associated with buffet onset are extracted from the steady base flow and subsequently integrated into the hidden layer of the neural network to impose physical constraints. With the test cases of the NACA0012 airfoil at various Mach numbers, the FINN model can accurately predict the damping representing the unsteady instability margin. Compared to the direct mapping input-output neural network (NN) model, the proposed method with shock wave feature-informed has enhanced accuracy, with an average relative error decreased by 70% at extrapolated Mach numbers. This research demonstrates the effectiveness of the FINN model in predicting the buffet onset, which leverages physics features derived from the more economical steady solution far from the onset boundary at a given predicted Mach number.

Statistical estimation of full-sky radio maps from 21cm array visibility data using Gaussian Constrained Realisations

Abstract An important application of next-generation wide-field radio interferometers is making high dynamic range maps of radio emission. Traditional deconvolution methods like CLEAN can give poor recovery of diffuse structure, prompting the development of wide-field alternatives like Direct Optimal Mapping and m-mode analysis. In this paper, we propose an alternative Bayesian method to infer the coefficients of a full-sky spherical harmonic basis for a drift-scan telescope with potentially thousands of baselines, that can precisely encode the uncertainties and correlations between the parameters used to build the recovered image. We use Gaussian Constrained Realisations (GCR) to efficiently draw samples of the spherical harmonic coefficients, despite the very large parameter space and extensive sky-regions of missing data. Each GCR solution provides a complete, statistically-consistent gap-free realisation of a full-sky map conditioned on the available data, even when the interferometer’s field of view is small. Many realisations can be generated and used for further analysis and robust propagation of statistical uncertainties. In this paper, we present the mathematical formalism of the spherical harmonic GCR-method for radio interferometers. We focus on the recovery of diffuse emission as a use case, along with validation of the method against simulations with a known diffuse emission component.

A novel approach to improve population mapping considering facility-based service capacity and land livability

Social sensing data, including points-of-interest (POI) and mobile position, are important data sources for population mapping. However, existing studies using POI data disregard the heterogeneity in facility size among the same type of POI and urban–rural differences. Moreover, mobile position data face biased issues. This study presents a hybrid model that considers facility-based service capacity (FSC) and land livability to map fine-scale population distributions. Based on extracting the FSC index by integrating POI, mobile phone positioning (MPP), and road network data, the district-level census population was disaggregated into 100-m grids using a random forest model. Subsequently, a regression equation was developed from the land use data to correct the estimated residuals. The results showed that the hybrid model exhibited considerably smaller errors than those of the POI density-based method and direct mapping of MPP (RMSE = 5,393.31, 7,348.91, and 7,824.76, respectively), effectively reducing population misallocation in extreme-density regions. Compared to WorldPop and LandScan datasets and a comparative model, our method reduced the RMSE by 30–60%, proving the effectiveness of integrating various social sensing and remote sensing data for improving population mapping. This study improves on an existing method as a thought-provoking step toward advancement in fine-scale population mapping research.

Fine-grained Fidgety Movement Classification using Active Learning.

Typically developing infants, between the corrected age of 9-20 weeks, produce fidgety movements. These movements can be identified with the General Movement Assessment, but their identification requires trained professionals to conduct the assessment from video recordings. Since trained professionals are expensive and their demand may be higher than their availability, computer vision-based solutions have been developed to assist practitioners. However, most solutions to date treat the problem as a direct mapping from video to infant status, without modeling fidgety movements throughout the video. To address that, we propose to directly model infants' short movements and classify them as fidgety or non-fidgety. In this way, we model the explanatory factor behind the infant's status and improve model interpretability. The issue with our proposal is that labels for an infant's short movements are not available, which precludes us to train such a model. We overcome this issue with active learning. Active learning is a framework that minimizes the amount of labeled data required to train a model, by only labeling examples that are considered "informative" to the model. The assumption is that a model trained on informative examples reaches a higher performance level than a model trained with randomly selected examples. We validate our framework by modeling the movements of infants' hips on two representative cohorts: typically developing and at-risk infants. Our results show that active learning is suitable to our problem and that it works adequately even when the models are trained with labels provided by a novice annotator.

Learning Chinese as a Second Language: Implications of the Character-Word Dual Function Model

Learning new words is fundamental in both first and second-language reading. There are, however, divided opinions on the best instructional approaches. Two widely used approaches across languages are whole-word focus and word-constituent focus. The appropriateness of each approach has varied historically, even within a single language (e.g., the debate between whole-word instruction and phonics in English). In teaching Chinese, both approaches are applied but to different learner groups. Whole-word instruction predominates in teaching Chinese as a second language (L2), while instruction for Chinese children focuses more on character-level mappings. It may seem reasonable in L2 Chinese instructions to focus on direct mappings between Chinese words and their L1 equivalent words. However, this raises a question: Is whole-word instruction the most efficient approach in L2 Chinese instruction? Based on an analysis of the Chinese writing system, we proposed a Character-Word Dual Function model of Chinese and tested its application of a dual-focus approach on both characters and words in L2 Chinese classroom instruction. Empirical findings support the advantage of this new approach compared to conventional whole-word instruction. We discuss the alignment between our findings and the Unified Computational Model and its implications for word instruction across languages.

IDENTIFICATION OF AQUIFERS BASED ON THE VERTICAL ELECTRICAL SOUNDING (VES) METHOD SCHLUMBERGER CONFIGURATION CASE STUDY: PULAU BAAI KAMPUNG MELAYU SUB-DISTRICT, BENGKULU CITY, INDONESIA

Investigation of the groundwater potential in the Pulau Baai area, Kampung Melayu Sub-district, Bengkulu City, must be carried out in such a way that the activities and needs of the people in the area can be fulfilled and the needs of the population in the area can be met. This study aims to determine the status of groundwater using the Schlumberger configuration geoelectric method. Measurements were made using a resistivity meter, and the results for each configuration depended on changes in resistivity. Measurements for each configuration depend on changes in resistivity at depth, the vertical direction (sounding), and the lateral direction (mapping), so hydrogeological analysis in this activity aims to get the maximum use of groundwater / underground water in aquifers for raw water needs. The dominant rock structures in the study area are clay, alluvium, siltstone, and sandstone, as well as some rocks with suitable porosity and permeability as water carriers, such as sand and gravel. However, what appears to have considerable potential is that groundwater is found at depths of 4-53 meters in VES 1, VES 2, VES 3, VES 5, and VES 10. The results of the analysis show that the location of the Pulau Baai, Kampung Melayu Sub-district, Bengkulu City Priority Utilization Area is within the groundwater storage area, so it can be used to meet the raw water needs of the study area.

Advancing instantaneous detonation prediction of condensed energetic materials based on high-order compressible multiphase fluid dynamics

The instantaneous detonation model (IDM) is widely used in simulating underwater explosions due to its efficiency and ability to ignore the detonation reaction process. In this study, we propose a new IDM to predict the fluid structure in the detonation zone of an RS211 explosive charge. This model is based on high-order solutions provided by the detonation shock dynamics model, where the spatial term is discretized using fifth-order WENO reconstruction in characteristic space and Lax–Friedrichs’s splitting and the temporal terms are discretized using a third-order TVD Runge–Kutta scheme. The interface motion is captured using the level-set method combined with MGFM, and a programmed burn model is provided to describe the generation and propagation of the detonation wave. The self-similarity of detonation wave propagation is validated, and the quantitative calculation formula of the instantaneous detonation model is obtained by averaging or curve fitting the dimensionless results. Consequently, the IDM of the RS211 charge is established using high-order polynomial approximations of the Taylor rarefaction zone and a constant static zone for 1D planar, cylindrical, and spherical RS211 charges. The application of the IDM involves direct mapping from the radial direction to the spatial structured grid for 1D planar, 2D cylindrical, and 3D spherical charges. Numerical results demonstrate that the IDM proposed in this paper shows good accuracy and high computational efficiency.

Robust flat bands in twisted trilayer graphene moiré quasicrystals

Moiré structures formed by twisting three layers of graphene with two independent twist angles present an ideal platform for studying correlated quantum phenomena, as an infinite set of angle pairs is predicted to exhibit flat bands. Moreover, the two mutually incommensurate moiré patterns among the twisted trilayer graphene (TTG) can form highly tunable moiré quasicrystals. This enables us to extend correlated physics in periodic moiré crystals to quasiperiodic systems. However, direct local characterization of the structure of the moiré quasicrystals and of the resulting flat bands are still lacking, which is crucial to fundamental understanding and control of the correlated moiré physics. Here, we demonstrate the existence of flat bands in a series of TTGs with various twist angle pairs and show that the TTGs with different magic angle pairs are strikingly dissimilar in their atomic and electronic structures. The lattice relaxation and the interference between moiré patterns are highly dependent on the twist angles. Our direct spatial mappings, supported by theoretical calculations, reveal that the localization of the flat bands exhibits distinct symmetries in different regions of the moiré quasicrystals.

Direct mapping of tyrosine sulfation states in native peptides by nanopore.

Sulfation is considered the most prevalent post-translational modification (PTM) on tyrosine; however, its importance is frequently undervalued due to difficulties in direct and unambiguous determination from phosphorylation. Here we present a sequence-independent strategy to directly map and quantify the tyrosine sulfation states in universal native peptides using an engineered protein nanopore. Molecular dynamics simulations and nanopore mutations reveal specific interactions between tyrosine sulfation and the engineered nanopore, dominating identification across diverse peptide sequences. We show a nanopore framework to discover tyrosine sulfation in unknown peptide fragments digested from a native protein and determine the sequence of the sulfated fragment based on current blockade enhancement induced by sulfation. Moreover, our method allows direct observation of peptide sulfation in ultra-low abundance, down to 1%, and distinguishes it from isobaric phosphorylation. This sequence-independent strategy suggests the potential of nanopore to explore specific PTMs in real-life samples and at the omics level.

A graph network-based learnable simulator for spatial-temporal prediction of rigid projectile penetration

Predicting plate penetration by rigid projectiles (PPRP) is crucial in terminal ballistics, with broad applications in civil and military engineering. Empirical and analytical methods face challenges in predicting field variables like displacement and stress in target plates. Although numerical methods offer high accuracy, they suffer from low computational efficiency. Herein, we introduce an efficient data-driven machine learning (ML) method based on graph neural networks (GNNs), named PGN, specifically tailored to address the PPRP problem. Unlike traditional ML methods that establish direct input-output mappings, PGN predicts comprehensive spatial-temporal information pertaining to the projectile-target interaction process. A thorough analysis of PGN's performance in terms of accuracy, computational efficiency and generalization ability was performed. Compared to validated results of numerical simulations, PGN maintained high precision with RMSE for displacement, stress, and strain predictions below 0.5 %, 9.5 %, and 2.1 %, respectively. It also achieved R2 values exceeding 0.92 for the time history of projectile velocity and acceleration, while requiring only 9.8 % of the computation time compared to LS-DYNA. In generalization tests, PGN exhibited remarkable adaptability in tackling challenging scenarios that extend far beyond the training data distribution, with overall RMSE between 11 % and 13 %. Furthermore, we find that the maximum information propagation capacity of a simulated physical system must meet or exceed the information propagation need of the real-world physical phenomenon it aims to replicate. Consequently, an approach was proposed to determine the critical connectivity radius of the massage passing method directly from the wave speed in the target medium, which greatly improved the accuracy and efficiency of PGN.

Bridge damage location and quantification under the moving vehicle loads based on deep learning multi-objective regression

The extraction of damage-sensitive features based on deep learning for the vibration response caused by vehicle-bridge interaction (VBI) has become a mainstream method in damage identification. However, most studies remain at the stage of qualitative damage assessment, not achieving precise numerical quantification of damage. They also typically involve multiple frequent runs using only a single dedicated vehicle. In this study, utilizing a deep neural network designed for multi-objective regression tasks, a novel method for bridge damage localization and quantification is presented with a multi-objective regressor designed to create a direct mapping between the curve of bridge deflection in time domain and the damage information matrix. The final quantitative damage values are obtained through statistical analysis. The method is evaluated on a simulated data set generated from VBI simulation using various 3D heavy vehicle models. The results demonstrate that this method can accurately locate and quantify damage even with unseen vehicle load conditions, damage scenarios, and measurement noise. The structure and depth affect the effectiveness of extracting damage-sensitive features from time-domain deflection during training. The study shows potential for real-time damage identification under normal operating conditions of real bridges in the rapid development of intelligent transportation systems.