Radon-full-waveform inversion for suppressing scalp reverberation and skull-induced aberration in transcranial photoacoustic computed tomography.

An efficient three-term conjugate gradient method with adaptive regularization and its application in image restoration

Entropy-augmented deep reinforcement learning with adaptive exploration for integrated energy and motor thermal management in hybrid electric vehicles

Physics-guided approaches offer a promising path toward accurate and generalisable impact identification in composite structures, especially when experimental data are sparse. This paper presents a hybrid framework for impact localisation and force estimation in composite plates, combining a data-driven implementation of First-Order Shear Deformation Theory (FSDT) with machine learning and uncertainty quantification. The structural configuration and material properties are inferred from dispersion relations, while boundary conditions are identified via modal characteristics to construct a low-fidelity but physically consistent FSDT model. This model enables physics-informed data augmentation for extrapolative localisation using supervised learning. Simultaneously, an adaptive regularisation scheme derived from the same model improves the robustness of impact force reconstruction. The framework also accounts for uncertainty by propagating localisation uncertainty through the force estimation process, producing probabilistic outputs. Validation on composite plate experiments confirms the framework's accuracy, robustness, and efficiency in reducing dependence on large training datasets. The proposed method offers a scalable and transferable solution for impact monitoring and structural health management in composite aerostructures.

Adaptive regularization for virtual sensing of structural deformation

Proximal Policy Optimization (PPO) is widely adopted for robotic continuous control, yet it can suffer from insufficient exploration and unstable policy updates in high-dimensional action spaces. This paper proposes Adaptive Exploration Proximal Policy Optimization (AE-PPO), an enhanced PPO framework that integrates (i) adaptive clipping, which adjusts the clipping range according to the observed magnitude of policy updates to better balance stability and learning progress, (ii) adaptive entropy regularization, which schedules the entropy weight across training to maintain effective exploration while avoiding excessive randomness. AE-PPO is evaluated on standard MuJoCo continuous control benchmarks (e.g., Walker2d, HalfCheetah, and Humanoid) and compared with PPO and representative baselines such as Trust Region Policy Optimization (TRPO) and Soft Actor Critic (SAC). The results show that AE-PPO achieves faster convergence and an improved final performance with reduced training variance, demonstrating more stable and efficient learning in challenging high-dimensional tasks.

Accurate electric vehicle (EV) charging load forecasting is essential for grid planning and resource allocation, yet existing approaches struggle with the inherent sparsity of charging data—a phenomenon characterized by excessive zeros representing periods of no charging activity. This paper addresses this challenge through a novel framework combining a Zero-Inflated Neural Network (ZINN) architecture with an Evolutionary Neural Architecture Search (ENAS) algorithm. ZINN explicitly decomposes the forecasting problem into binary classification (predicting charging occurrence) and regression (estimating energy magnitude conditioned on occurrence), enabling the model to learn distinct patterns for the absence and presence of charging events. Rather than relying on manually designed architectures, ENAS automatically discovers optimal encoder and decoder configurations from a comprehensive search space encompassing modern architectures (LSTM, GRU, Transformer, and iTransformer), layer configurations, activation functions, and hyperparameters. The evolutionary algorithm balances prediction accuracy with computational efficiency through multi-objective optimization. Extensive experiments on real-world EV charging data from 30 stations in Wuhan demonstrate that the ZINN+ENAS framework achieves the lowest prediction error compared to conventional baselines, with the discovered optimal configuration substantially outperforming hand-crafted designs. Comprehensive ablation studies reveal that the asymmetric dual-head architecture and adaptive regularization strategies are critical for handling data sparsity. These findings highlight the importance of explicit zero-inflation modeling and automated architecture discovery for specialized forecasting tasks, providing practitioners with an open-source framework for practical EV charging load prediction.

{A novel Bayesian approach using horseshoe priors is developed for exponential random graph models (ERGMs), addressing degeneracy and improving inference. Simulation studies demonstrate superior performance over classical ERGMs.} Exponential Random Graph Models (ERGMs) are widely used for analyzing network data but often face challenges such as parameter estimation difficulties and model degeneracy. This paper introduces the Bayesian Horseshoe Prior ERGM (BHSERGM), a hierarchical Bayesian framework that leverages the adaptive shrinkage of the horseshoe prior to improve inference in sparse, high-dimensional networks. Implemented via a parallel adaptive direction sampling algorithm, BHSERGM enhances computational efficiency and convergence. Simulation studies and empirical applications demonstrate that BHSERGM shows improved estimation accuracy and enhanced recovery of network structure compared to conventional Bayesian ERGMs, while its adaptive regularization approach appears effective in mitigating degeneracy issues commonly encountered in standard ERGM frameworks.

GGAR: gradient guided adaptive regularization enhances deep learning classification of brassica species using codon usage bias.

Satellite-based precipitation estimates (SPE) provide essential spatial coverage and near real-time availability for hydrological applications but often exhibit systematic biases in regions characterized by complex terrain and strong climatic variability, limiting their reliability for flood-related studies. To address these limitations, this study proposes an Adaptive Regularization framework integrated within a Long Short-Term Memory (LSTM) model to enhance satellite–gauge rainfall fusion beyond conventional optimization strategies. The framework dynamically adjusts learning rate and weight decay during training based on validation performance and overfitting indicators, improving training stability, data efficiency, and model generalization across diverse precipitation regimes. The proposed approach was applied to refine Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG-Final) daily rainfall estimates over the flood-prone east coast of Peninsular Malaysia. Model performance was assessed against ten optimization algorithms using correlation coefficient (CC), mean absolute error (MAE), normalized root mean squared error (NRMSE), percentage bias (PBias), and Kling–Gupta efficiency (KGE). Results show that the Adaptive Regularization framework consistently outperforms all benchmark optimizers, achieving an MAE of 6.87, CC of 0.68, NRMSE of 1.84, and KGE of 0.56. Overall, the proposed framework enhances spatial consistency and robustness across monsoon seasons, offering a scalable solution for improving SPE in flood-prone regions.

A 3D inversion method of TEM combining PSO-NLCG optimization and adaptive regularization.

Accurate feature selection is critical for machine learning in medical diagnosis, yet conventional methods often fail to capture complex non-linear relationships in biomedical data. This study introduces an advanced feature selection approach that integrates distance correlation with adaptive weighting to enhance cardiac arrhythmia detection. The proposed method ranks features based on distance correlation, applies an inverse penalty weighting scheme to suppress highly correlated features while emphasizing moderately correlated ones, and incorporates RBF kernel transformation followed by LASSO refinement. Fifteen feature selection techniques were evaluated on an electrocardiographic database of 279 morphological and physiological features using 4-fold cross-validation with a neural network classifier. The proposed method outperformed all alternatives, including the best conventional approach, by effectively capturing non-linear dependencies, mitigating multicollinearity and overfitting, and leveraging synergistic kernel-based interaction modeling with sparse selection. These results demonstrate that combining statistical dependence measures, adaptive regularization, and non-linear transformations provides a robust framework for feature selection in cardiac arrhythmia classification and broader medical informatics applications.

Recent advancements in single-cell multi-omics technologies have significantly improved our ability to explore cellular heterogeneity at an unprecedented resolution. These innovations enable the simultaneous profiling of genomic, transcriptomic, proteomic, and epigenetic data at the single-cell level, providing comprehensive insights into cellular states and their regulatory mechanisms. However, integrating multi-omics data remains challenging due to its high dimensionality, technical noise, and biological complexity. To address these challenges, we introduce scWDAC (single-cell weighted distance adaptive clustering), a novel clustering method for single-cell multi-omics data. scWDAC utilizes a weighted distance penalty and adaptive graph regularization to effectively integrate multiple omics layers. Key innovations of scWDAC include using a weighted distance penalty to capture cell-to-cell similarities across different omics modalities, and applying adaptive graph regularization on a consensus matrix to enforce cross-modal consistency. The framework optimizes both global consistency and local accuracy, ensuring a robust exploration of cellular structures across all omics layers. The effectiveness of scWDAC is evaluated through extensive experiments on ten paired single-cell multi-omics datasets. The results demonstrate that scWDAC outperforms existing clustering methods in terms of clustering accuracy, robustness to noise, and biological interpretability.

NLMap-ATVR: A novel combination of nonlinear mapping network and adaptive total variation regularization for MRI denoising.

Adaptive regularization weight selection for compressed sensing MRI reconstruction.

Towards heterogeneity-aware federated self-supervised learning via knowledge anchoring

This article is motivated by the challenge of identifying interpretable brain functional connectivity biomarkers for Alzheimer's disease (AD) progression using high-dimensional functional magnetic resonance imaging (fMRI) data, where predictors comprise both heterogeneous disease-stage activation patterns and strongly similar functional connections. We propose a unified statistical method, called Sparse Multi-task Adaptive Regularization Truncation (SMART), to simultaneously address three critical challenges: (1) High dimensionality is resolved through an -penalty ( ) that selects sparse, stage-distinct functional connections; (2) Disease-stage heterogeneity is accommodated via an -penalty ( ) that maintains stable activation patterns across tasks; (3) Connection collinearity is mitigated using a truncated penalty (TLP; ) that adaptively groups edges with similar cross-task profiles without pre-specified structure. SMART offers key advantages over existing methods: Its joint regularization naturally handles smooth activation patterns across stages, while the TLP's dual parameters ( for adaptive grouping threshold, for sparsity control) provide a principled trade-off between biological fidelity and model complexity. Computationally, we develop a DC-ADMM algorithm that transforms the optimization into tractable subproblems, establishing finite-step convergence to KKT points. Comprehensive simulation studies and real-data analysis of AD neuroimaging data demonstrate SMART's superior accuracy in connectivity biomarker identification, enhanced stability in feature selection, and improved interpretability for AD cohort studies. An accompanying R package, SMART, is available on GitHub ( https://github.com/LDstat/SMART).

Efficient Implementation of Third-order Tensor Methods with Adaptive Regularization for Unconstrained Optimization

This paper evaluates an Adaptive LASSO-MGARCH model for multivariate volatility forecasting, with applications in green and conventional bonds, equities, energy commodities, and EU carbon allowances. By introducing coefficient-specific adaptive penalisation directly into the multivariate GARCH variance equations, the model delivers a sparse and data-driven volatility spillover structure while preserving the positive definiteness of the conditional covariance matrix. Using daily data on green and conventional bonds, equities, energy commodities, and carbon allowances, we show that adaptive regularisation substantially reduces model complexity and improves economic interpretability relative to an unpenalised MGARCH benchmark. Out-of-sample forecasting experiments at multiple horizons demonstrate that the Adaptive LASSO-MGARCH model consistently achieves lower covariance forecast losses, and statistical tests based on the White reality check confirm that these improvements are significant across alternative loss functions.

Lightweight generative channel estimation with adaptive regularization in massive MIMO systems