Multimodal approaches that integrate protein structure and sequence have achieved remarkable success in protein-protein interface prediction. However, extending these methods to protein-peptide interactions remains challenging due to the inherent conformational flexibility of peptides and the limited availability of structural data that hinders direct training of structure-aware models. To address these limitations, we introduce GeoPep, a novel framework for peptide binding site prediction that leverages transfer learning from ESM3, a multimodal protein foundation model. GeoPep fine-tunes ESM3's rich prelearned representations from protein-protein binding to address the limited availability of protein-peptide binding data. The fine-tuned model is further integrated with a Kolmogorov-Arnold Network (KAN)-based architecture for complex nonlinear approximation. Furthermore, the model is trained using distance-based loss functions that exploit 3D structural information to enhance binding site prediction. Comprehensive evaluations demonstrate that GeoPep significantly outperforms existing methods in protein-peptide binding site prediction by effectively capturing sparse and heterogeneous binding patterns.

Out-of-Distribution (OoD) detection is vital for the reliability of deep neural networks, the key of which lies in effectively characterizing the disparities between OoD and In Distribution (InD) data. In this work, such disparities are exploited through a fresh perspective of non-linear feature sub spaces. That is, a discriminative non-linear subspace is learned from InD features to capture representative patterns of InD, while informative patterns of OoD features cannot be well captured in such a subspace due to their different distribution. Grounded on this perspective, we exploit the deviations of InD and OoD features in such a non-linear subspace for effective OoD detection. To be specific, we leverage the framework of Kernel Principal Component Analysis (KPCA) to attain the discriminative non linear subspace and deploy the reconstruction error on such subspace to distinguish InD and OoD data. Two challenges emerge: (i) the learning of an effective non-linear subspace, i.e., the selection of kernel function in KPCA, and (ii) the computation of the kernel matrix with large-scale InD data. For the former, we reveal two vital non-linear patterns that closely relate to the InD-OoD disparity, leading to the establishment of a Cosine Gaussian kernel for constructing the subspace. For the latter, we introduce two techniques to approximate the Cosine-Gaussian kernel with significantly cheap computations. In particular, our approximation is further tailored by incorporating the InD data confidence, which is demonstrated to promote the learning of discriminative subspaces for OoD data. Our study presents new insights into the non-linear feature subspace for OoD detection and contributes practical explorations on the associated kernel design and efficient computations, yielding a KPCA detection framework with distinctively improved efficacy and efficiency.

Model predictive control (MPC) for aeroengines requires accurate prediction of complex nonlinear dynamics, which is challenging to achieve using traditional modeling approaches. While neural networks (NNs) offer strong nonlinear approximation capability, their embedded deployment suffers from approximation and quantization errors, leading to persistent steady-state offsets in closed-loop control. To address these limitations, this article proposes a real-time offset-free MPC framework that incorporates an NN-based prediction model with a multivariable adaptive error compensator to eliminate steady-state deviations and enhance tracking performance. Moreover, to ensure real-time feasibility, a low-rank approximation linearization method is developed to reduce computational complexity, and the entire framework is hardware-accelerated on a Zynq deep-learning processing unit (DPU) for embedded execution. Finally, hardware-in-the-loop (HIL) experiments on realistic aeroengine control scenarios demonstrate that the proposed approach achieves a faster dynamic response, improved steady-state accuracy, and a lower computation latency compared with conventional MPC, confirming its potential for practical aerospace applications.

This study is conducted for the comparative analysis of unsteady magnetohydrodynamic flow and heat-mass transfer behaviour of two different hybrid nanofluids such as TiO2, Cu /Water and TiO2, Cu /Water + Ethylene glycol(50:50) over an inclined vertical porous plate subjected to uniform angled magnetic strength, thermal radiation, heat source/sink and coupled influence of Soret and Dufour effects. The flow is induced by buoyancy force, which is modelled using the nonlinear Boussinesq approximation. The equations that can govern the system are non-dimensionalized by using appropriate similarity transformations and dimensionless parameters. Multi regular Perturbation method is employed to get the solution. The various pertinent parameters involved in the system are considered for the analysis of their implications on the various profiles. The analysis of this study reveals that the velocity of the fluid is enhanced with higher values of permeability, buoyancy, Dufour and Soret parameters, whereas diminished with high values of magnetic field strength, magnetic field inclination, volume fraction of the nanoparticles and plate angle parameters. As the values of radiation, Dufour and volume fraction parameters increased, so does the temperature profile. Concentration profile is developed with high Soret number value. Moreover, the numerical values of engineering quantities, including the Skin friction coefficient and Nusselt number are presented in tabular form. The findings of this study are having applications relevant to microelectronics cooling, heat exchangers and battery thermal management systems.

Agricultural tractor traction performance prediction is crucial for reducing fiel consumption and improving operation quality in modern agriculture. To address the limitations of traditional empirical modeling and regression methods in predicting tractor traction performance, where model accuracy and computational efficiency are often constrained by multidimensional nonlinear characteristics, this study introduces a Physics-informed Neural Network (PINN) framework. The proposed approach integrates the nonlinear approximation capability of neural networks with prior knowledge from traction dynamics, and incorporates physics-based constraint terms into the loss function to enhance physical consistency and generalization. Unlike conventional black-box data-driven models, the PINN framework simultaneously fits observed data and enforces consistency with physical laws during training, which prevents physically implausible predictions. Experimental results show that the PINN achieves high accuracy, particularly in predicting traction power and fuel consumption rate, with an average coefficient of determination (R2) above 95% across four output indicators. In addition, the predicted trends are highly consistent with measured data, demonstrating the potential of this method to provide strong technical support for the optimization and design of agricultural machinery power systems.

Mathematical modeling plays a key role in understanding and optimizing transport system operations under uncertain and dynamic conditions. This study proposes a data-driven predictive framework for estimating passenger-accepted vehicle occupancy, addressing a critical gap in transport system planning under public health-related constraints. Using data from a structured survey conducted across seven Southeast European countries (N = 476), the study integrates statistical analysis and machine learning approaches to model acceptable occupancy levels across multiple transport modes, including passenger cars, taxis, tourist buses, and public buses. The problem is formulated as a predictive mapping between multidimensional input variables and occupancy acceptance levels, modeled using both probabilistic and nonlinear function approximation methods. The results highlight that age, gender, and area of residence are the most significant determinants of occupancy acceptance, while education level has limited predictive relevance. Furthermore, a multi-layer feedforward artificial neural network is developed to capture nonlinear relationships between variables, achieving strong predictive performance (minimum MSE = 0.0089). The main contribution of this research lies in linking behavioral data with predictive modeling to quantify acceptable occupancy thresholds and support realistic simulation of passenger responses in crisis conditions. The proposed modeling framework contributes to transport system planning, enabling data-driven capacity management, enhanced safety strategies, and improved resilience of passenger transport operations.

Robust adaptive H∞ fault-tolerant predictive control for air-ground integrated highway emergency self-organizing network systems based on RBF-DNN approximation.

• A hybrid CFD-DNN framework predicts ship motions in irregular waves using residual learning. • The proposed strategy explicitly corrects systematic numerical dissipation and phase lags in RANS simulations. • Swish activation and SmoothL1 loss functions enable high accuracy despite sparse training datasets. • The method reduces computational costs from days to milliseconds to facilitate real-time digital twinning. Accurate time-domain prediction of ship motions in irregular waves is critical for assessing dynamic stability and operability. While Reynolds-Averaged Navier-Stokes (RANS) simulations provide high-fidelity insights, they are constrained by prohibitive computational costs and often exhibit systematic discrepancies, such as numerical dissipation and phase lags, under nonlinear conditions. This paper proposes a data-efficient hybrid modeling framework that synergizes the physical priors of CFD with the nonlinear approximation capability of Deep Neural Networks (DNN) to enhance prediction accuracy. Unlike traditional surrogate models, the proposed architecture employs a residual learning strategy, explicitly training the network to correct the systematic errors between numerical predictions and model test measurements. To address the challenges of data sparsity and experimental noise common in marine engineering, the network incorporates Swish activation functions and a robust SmoothL1 loss function. The framework is validated using experimental data from a Chemical Tanker in irregular waves. Despite being trained on a limited dataset (720 samples), the hybrid model significantly outperforms standalone CFD. Statistical analysis shows a reduction in Mean Absolute Error (MAE) for heave motion from 0.97 m to 0.59 m. The proposed approach reduces computational time from days to milliseconds while maintaining physical consistency, offering a robust tool for digital twinning and rapid design evaluation under data-scarce conditions.

A simple computational model is proposed for investigating the dynamic behavior of an ornithopter with elastic flapping wings of high aspect ratio. The wings are modeled as an elongated, orthotropic composite plate of a rod-like type. It is assumed that the orthotropy axes of the plate material do not coincide with the axes of the chosen Cartesian coordinate system for the wing, which allows for the description of its coupled bending-torsional vibrations. The ornithopter’s body core is modeled as an absolutely rigid solid body. The developed deformation model for the wing is based on the relations of the refined shear model of S.P. Timoshenko, formulated for rods in a geometrically nonlinear approximation, neglecting compression in the transverse directions. The kinematic coupling conditions between the wings and the body core are formulated. Using these conditions and based on the variational D’Alembert-Lagrange principle, the corresponding equilibrium (motion) equations and boundary conditions are derived for the considered ornithopter elements. The force coupling conditions at the wing-body junction are also obtained, which essentially represent the equations of perturbed motion for the body core.

Accurate demand forecasting is critical for firm-level operational and strategic decisions, yet conventional econometric and machine learning models largely abstract from systematic behavioral distortions in consumer decision-making. Drawing on behavioral demand theory, this study develops and empirically evaluates a hybrid forecasting framework that integrates reference dependence, habit persistence, and attention-based mechanisms into econometric, machine learning, and ensemble demand forecasting models. Using firm-level demand data and a rolling-origin validation design, we compare traditional baseline models with behaviorally augmented specifications across multiple forecast horizons and error metrics. The results show that models incorporating behavioral variables consistently and significantly outperform standard econometric and machine learning benchmarks out of sample. Reference price losses exert a substantially stronger predictive influence than gains, consistent with loss aversion, while habit persistence dominates short-horizon forecasts and attention and sentiment measures contribute most at medium horizons. Further gains are achieved through forecast ensembles that combine behaviorally augmented econometric and machine learning models, indicating complementary strengths in structural discipline and non-linear approximation. These findings demonstrate that behavioral demand mechanisms are not only explanatory but also predictively relevant at the firm level. The study contributes to behavioral demand theory by extending it into an explicitly predictive context, advances forecasting methodology by showing the value of theory-guided feature augmentation, and offers a scalable framework for firms seeking more accurate and behaviorally informed demand forecasts. Keywords: Behavioral Demand, Econometric Models, Forecast Ensembles, Habit Persistence, Loss Aversion.

This article explores global consensus tracking control for multiagent systems with unknown time-varying gains and nonlinearities, subject to quantitative performance constraints. The challenge lies in designing a controller that achieves global consensus without relying on the system's initial conditions while also ensuring the prescribed settling time and the convergence accuracy of synchronization errors. First, a concise, differentiable, piecewise continuous regulation function is proposed. The combination of this regulation function with error transformation addresses the singularity issue associated with initial conditions. Then, a piecewise performance function is also introduced to quantify both settling time and steady-state accuracy of synchronization errors. Integrating the regulation and performance functions yields a novel, low-complexity, robust method for enforcing quantitative performance constraints. This approach ensures global consensus under specified constraints and eliminates the need for nonlinear function approximation, parameter estimation, higher order derivative computation, or adaptive law design. Finally, the effectiveness of the proposed method is demonstrated through comparative simulations involving a planar robotic system.

Joint probability aggregation for regional wind power forecasting via R-vine copula and Kolmogorov-Arnold networks

Interval type-2 fuzzy reservoir cerebellar model articulation controller design for antilock braking systems

Accurate drug-target affinity (DTA) prediction is critical for drug discovery and repurposing. However, existing models often struggle with generalizing to unseen drug-target pairs, lack interpretability, and fail to integrate heterogeneous biological features effectively. To overcome these challenges, we introduce KANPM-DTA, a deep learning framework designed to capture richer biochemical interactions and improve prediction reliability. Specifically, an ESM-guided protein graph construction strategy incorporates evolutionary and structural information to overcome underexplored protein representations. A gated fusion mechanism was employed to integrate drug-protein graph features, while linear attention captures cross-modal dependencies that enhance discriminative power. For the final affinity prediction, a Kolmogorov-Arnold network was used, offering a stronger nonlinear approximation and improved interpretability. Comprehensive experiments on benchmark datasets demonstrate that KANPM-DTA significantly outperforms state-of-the-art methods. On the Davis, KIBA, Metz, and BindingDB datasets, we achieved significant performance improvements under warm setting, with MSE reductions of 6.42%, 4.86%, 4.44%, and 5.46%, CI increases of 0.45%, 0.34%, 0.48%, and 0.80%, and $r_{m}^{2}$ gains of 1.85%, 0.90%, 0.84%, and 1.05%, respectively. Moreover, a case study on the epidermal growth factor receptor further highlights the effectiveness of KANPM-DTA in predicting DTAs for unknown drug-target pairs, emphasizing its potential for real-world applications in drug discovery. However, wet-lab validation is required to assess the applicability of the results.

Nonlinear approximation of harmonic functions from shifts of the Newtonian kernel in BMO

The Taylor approximation theorem is a fundamental tool in numerical analysis, providing a local polynomial representation of smooth functions. In practical computations, a function f is approximated by a finite Taylor polynomial Pn, and controlling the resulting truncation error is of central importance. In this paper, we introduce two novel a posteriori error estimation techniques for Taylor polynomial approximations. The proposed estimators are fully computable and do not require prior bounds on the (n+1)st derivatives of f. We prove that the estimators converge to the exact error both pointwise and in the L2-norm as n→∞, and we establish their asymptotic sharpness through effectivity analysis. Based on these results, we develop two adaptive algorithms that automatically determine the minimal degree n required to achieve a prescribed tolerance, either at a specific point or over a domain. We further extend the analysis to multivariate functions and show that analogous estimators and effectivity properties hold in higher dimensions. Numerical experiments are presented to validate the theoretical results and demonstrate the practical performance of the proposed methods.

This paper describes a high-order strongly nonlinear (SNL) model for long waves in the presence of a variable bottom, which is a generalisation of the model for a flat bottom (Choi 2022 a, J. Fluid Mech. vol. 945, A15). This asymptotic model written in terms of the bottom velocity is obtained using systematic expansion with a single small parameter measuring the ratio of the water depth to the characteristic wavelength and is found linearly stable at any order of approximation. To test the high-order SNL model with a variable bottom, we solve numerically the first- and second-order models using a pseudo-spectral method to study the deformation or generation of long waves over a variable bottom. Specifically, we consider two examples: (i) the propagation of cnoidal waves over a fixed bottom topography, and (ii) the forced generation of solitary waves by a submerged topography moving steadily with a transcritical speed. The computed results are then compared with the fully nonlinear computation using a boundary integral method as well as the numerical solutions of the weakly nonlinear long wave model. It is found that the second-order SNL model for the bottom velocity is suitable for stable numerical computations and produces accurate solutions even for a relatively large-amplitude initial wave or submerged topography.

Ecarin-based coagulation monitoring of argatroban in patients with heparin-induced thrombocytopenia: a prospective observational study.

Abstract This paper proposes a sliding-mode-based optimal attitude control framework that integrates reinforcement learning (RL) and sliding mode control (SMC) to address model uncertainties and unknown time-varying disturbances in quadrotor UAVs. The SMC is embedded into the optimal control design to achieve coordinated regulation of multiple attitude states, thereby enhancing closed-loop robustness and fast convergence performance. A neural network is introduced to perform online approximation and adaptive compensation of unknown nonlinearities and unknown time-varying disturbances in the UAV attitude dynamics, which reduces the dependence on an accurate mathematical model and improves control accuracy. An actor–critic reinforcement learning architecture is adopted to enable online optimization of the attitude control policy without requiring persistent excitation or continuous reward conditions, allowing the adaptive parameters to be effectively trained. Furthermore, the stability of the entire control system is rigorously analyzed using Lyapunov theory, guaranteeing that the attitude tracking errors are semi-globally uniformly ultimately bounded (SGUUB). Comprehensive numerical simulations and real-time flight experiments, including comparative studies with existing control strategies, are conducted to validate the effectiveness, robustness, and practical feasibility of the proposed method. The results demonstrate that the proposed control framework provides improved adaptability, control accuracy, and engineering applicability for quadrotor UAVs operating in complex and uncertain environments.

Structural Health Monitoring (SHM) aims to monitor in real-time the health state of engineering structures. For thin structures, Lamb Waves (LW) are very efficient for SHM purposes. A bonded piezoelectric transducer (PZT) emits LW in the structure in the form of a short tone burst. This initial wave packet (IWP) is then propagating in the structure and interacts with its boundaries and discontinuities, such as the presence of damage, generating additional waves packets. In this sense, both the geometry itself and the presence of damage can produce complex behaviors in the measured signal from sensors, which makes the extraction of features that will be used later to evaluate damage detection very complicated in complex scenarios. To solve this issue, here an innovative Deep Learning technique called Rank Reduction Autoencoder (RRAE) is considered. The RRAE consists on an autoencoder whose latent space is restricted to be expressed as a low-rank SVD approximation, capturing in this sense only the most important features of the studied signals, by keeping the advantage of learning complex behaviours by means of a nonlinear approximation. The novelty proposed in this work consists of adding an additional restriction in the latent space of the RRAE so that its content is as representative as possible to produce damage detection. This is achieved by means of a 2D Convolutional Neural Network (CNN) that takes as inputs the latent space and delivers as prediction the location of damage. Therefore, the training of the RRAE plus the extraction of important features by means of a MLP all together allows to obtain at convergence a powerful tool for damage detection in the SHM field. To illustrate the proposed technique, it is applied for the detection of damage on a thin plate.