arXiv subjects

Zhiwei Gao

Publications and source records attributed to Zhiwei Gao.

At least 19 recordsLinked to original sources

arXiv2026

What Does Activation Steering Control? Attribution Across Answer Encodings and Output-Sensitive Subspaces

Activation steering is often evaluated under the answer encoding used to construct the direction. A reported gain may reflect the intended judgment or compatibility with answer identifiers seen during construction. We introduce Cross-Encoding Steering Evaluation, which freezes an intervention while re-encoding answers to the same held-out items. On NormBank, after A/B/C identifiers are reassigned, contrastive activation addition (CAA) induces larger target-versus-source score changes for the extraction indices than for the semantic labels under the new mapping. We call this extraction-index following. Varying identifier vocabulary (A/B/C, X/Y/Z, or 1/2/3) and row order shows that the effect tracks extraction index rather than row position. After matching direction norms across layers, extraction-index following emerges mainly at later depths. A low-rank output-sensitive component containing 15.4% of the direction's squared norm retains 96.3% of this effect. An Inference-Time Intervention (ITI)-style method also favors extraction-index over semantic-label following on NormBank in three models. In aggregate, MNLI favors extraction-index following, whereas Social Chemistry 101 (SC101) favors semantic-label following. Multiple-choice and open-ended evaluations can yield different behavioral conclusions. Thus, a steering gain under one answer encoding does not by itself identify what the intervention controls.

cs.CL↗

arXiv2026

Spectral Audit of In-Context Operator Networks

Existing evaluations of neural operators and in-context operator learning rely primarily on prediction error, but accurate output prediction does not guarantee the correct local dynamical structure. A model may match solutions while exhibiting incorrect sensitivities, distorted frequency response, spurious mode coupling, or unstable tangent behavior. We introduce a Jacobian-based spectral audit for in-context operator learning. For a fixed prompt, we differentiate the network output with respect to the query function and view the resulting Jacobian as a learned tangent operator. Projecting it onto Fourier modes, we obtain a local spectral characterization of the inferred operator, including frequency-dependent gains, phase structure, and cross-mode coupling. The audit complements standard prediction metrics by testing whether the model reproduces local mechanisms of the underlying PDE operator rather than only outputs. Across benchmarks, the audit reveals distinct operator-level phenomena, including phase transport, viscosity-dependent damping, nonlinear mode coupling, and reaction--diffusion stability structure. It also detects failures partially hidden by prediction-error metrics, including high-frequency degradation, incorrect phase recovery, and prompt--operator inconsistencies. Corrupted or internally inconsistent prompts lead to degraded tangent-operator structure even when pointwise predictions remain partially accurate. Our results suggest that prediction accuracy and local operator fidelity are distinct properties of learned neural operators. Our framework also provides a diagnostic for stability, sensitivity, and operator consistency.

math.NA↗

arXiv2026

Proposal-Guided Greedy Surrogate Refinement for PDE-Driven High-Dimensional Rare-Event Estimation

Accurate surrogate construction for PDE-driven high-dimensional rare-event simulation is challenging when performance evaluations are expensive. Since a globally accurate surrogate may require many high-fidelity evaluations, adaptive importance sampling provides a natural localization tool: its evolving proposal distribution progressively identifies the failure-relevant region. Motivated by this observation, we propose a surrogate-assisted adaptive importance sampling framework that refines the surrogate locally along the evolving proposal, rather than over the entire input space. The surrogate combines an encoder with a neural network, providing a low-dimensional latent representation for both prediction and sample selection. At each adaptive iteration, candidates drawn from the current proposal are selected by a greedy latent-space rule balancing proximity to the estimated failure boundary and sample diversity. The selected samples are evaluated by the high-fidelity model and used to refine the surrogate, which then guides the subsequent cross-entropy-type adaptive proposal update. We establish one-step proposal stability bounds under local surrogate errors, together with surrogate-induced misclassification and finite-sample estimation error bounds. Numerical experiments on multimodal benchmarks and PDE-driven rare-event problems up to 100 dimensions show that the proposed method achieves accuracy comparable to true-model adaptive importance sampling while requiring substantially fewer high-fidelity evaluations.

math.NA↗

arXiv2026

Probabilistic Predictions of Process-Induced Deformation in Carbon/Epoxy Composites Using a Deep Operator Network

Fiber reinforcement and polymer matrix respond differently to manufacturing conditions due to mismatch in coefficient of thermal expansion and matrix shrinkage during curing of thermosets. These heterogeneities generate residual stresses over multiple length scales, whose partial release leads to process-induced deformation (PID), requiring accurate prediction and mitigation via optimized non-isothermal cure cycles. This study considers a unidirectional AS4 carbon fiber/amine bi-functional epoxy prepreg and models PID using a two-mechanism framework that accounts for thermal expansion/shrinkage and cure shrinkage. The model is validated against manufacturing trials to identify initial and boundary conditions, then used to generate PID responses for a diverse set of non-isothermal cure cycles (time-temperature profiles). Building on this physics-based foundation, we develop a data-driven surrogate based on Deep Operator Networks (DeepONets). A DeepONet is trained on a dataset combining high-fidelity simulations with targeted experimental measurements of PID. We extend this to a Feature-wise Linear Modulation (FiLM) DeepONet, where branch-network features are modulated by external parameters, including the initial degree of cure, enabling prediction of time histories of degree of cure, viscosity, and deformation. Because experimental data are available only at limited time instances (for example, final deformation), we use transfer learning: simulation-trained trunk and branch networks are fixed and only the final layer is updated using measured final deformation. Finally, we augment the framework with Ensemble Kalman Inversion (EKI) to quantify uncertainty under experimental conditions and to support optimization of cure schedules for reduced PID in composites.

cs.CE↗

arXiv2026

Nonlinear parametrization solver for fractional Burgers equations

Fractional Burgers equations pose substantial challenges for classical numerical methods due to the combined effects of nonlocality and shock-forming nonlinear dynamics. In particular, linear approximation frameworks-such as spectral, finite-difference, or discontinuous Galerkin methods-often suffer from Gibbs-type oscillations or require carefully tuned stabilization mechanisms, whose effectiveness degrades in transport-dominated and long-time integration regimes. In this work, we introduce a sequential-in-time nonlinear parametrization (STNP) for solving fractional Burgers equations, including models with a fractional Laplacian or with nonlocal nonlinear fluxes. The solution is represented by a nonlinear parametric ansatz, and the parameter evolution is obtained by projecting the governing dynamics onto the tangent space of the parameter manifold through a regularized least-squares formulation at each time step. This yields a well-posed and stable time-marching scheme that preserves causality and avoids global-in-time optimization. We provide a theoretical analysis of the resulting projected dynamics, including a stability estimate and an a posteriori error bound that explicitly decomposes the total error into contributions from initial condition fitting, projection residuals, and discretization of fractional operators. Our analysis clarifies the stabilizing role of regularization and quantifies its interaction with the nonlocal discretization error. Numerical experiments for both fractional Burgers models demonstrate that STNP achieves oscillation-free shock resolution and accurately captures long-time dynamics. The method consistently outperforms high-order spectral schemes augmented with spectral vanishing viscosity, while requiring significantly fewer degrees of freedom and avoiding ad hoc stabilization.

math.NA↗

arXiv2025

A Short Report on Importance Sampling for Rare Event Simulation in Diffusions

In this manuscript, we investigate importance sampling methods for rare-event simulation in diffusion processes. We show, from a large-deviation perspective, that the resulting importance sampling estimator is log-efficient. This connection is established via a stochastic optimal control formulation, and the associated Hamilton--Jacobi--Bellman (HJB) equation is derived using dynamic programming. To approximate the optimal control, we adopt a spectral parameterization and employ the cross-entropy method to estimate the parameters by solving a least-squares problem. Finally, we present a numerical example to validate the effectiveness of the cross-entropy approach and the efficiency of the resulting importance sampling estimator.

math.NA↗

arXiv2025

Multi-Modal Robust Enhancement for Coastal Water Segmentation: A Systematic HSV-Guided Framework

Coastal water segmentation from satellite imagery presents unique challenges due to complex spectral characteristics and irregular boundary patterns. Traditional RGB-based approaches often suffer from training instability and poor generalization in diverse maritime environments. This paper introduces a systematic robust enhancement framework, referred to as Robust U-Net, that leverages HSV color space supervision and multi-modal constraints for improved coastal water segmentation. Our approach integrates five synergistic components: HSV-guided color supervision, gradient-based coastline optimization, morphological post-processing, sea area cleanup, and connectivity control. Through comprehensive ablation studies, we demonstrate that HSV supervision provides the highest impact (0.85 influence score), while the complete framework achieves superior training stability (84\% variance reduction) and enhanced segmentation quality. Our method shows consistent improvements across multiple evaluation metrics while maintaining computational efficiency. For reproducibility, our training configurations and code are available here: https://github.com/UofgCoastline/ICASSP-2026-Robust-Unet.

cs.CV↗

arXiv2025

Safe cross-entropy-based importance sampling for rare event simulations

The Improved Cross-Entropy (ICE) method is a powerful tool for estimating failure probabilities in reliability analysis. Its core idea is to approximate the optimal importance-sampling density by minimizing the forward Kullback-Leibler divergence within a chosen parametric family-typically a mixture model. However, conventional mixtures are often light-tailed, which leads to slow convergence and instability when targeting very small failure probabilities. Moreover, selecting the number of mixture components in advance can be difficult and may undermine stability. To overcome these challenges, we adopt a weighted cross-entropy-penalized expectation-maximization (EM) algorithm that automatically prunes redundant components during the iterative process, making the approach more stable. Furthermore, we introduce a novel two-component mixture that pairs a light-tailed distribution with a heavy-tailed one, enabling more effective exploration of the tail region and thus accelerating convergence for extremely small failure probabilities. We call the resulting method Safe-ICE and assess it on a variety of test problems. Numerical results show that Safe-ICE not only converges more rapidly and yields more accurate failure-probability estimates than standard ICE, but also identifies the appropriate number of mixture components without manual tuning.

math.NA↗

arXiv2025

DeepSeek vs. ChatGPT vs. Claude: A Comparative Study for Scientific Computing and Scientific Machine Learning Tasks

Large Language Models (LLMs) have emerged as powerful tools for tackling a wide range of problems, including those in scientific computing, particularly in solving partial differential equations (PDEs). However, different models exhibit distinct strengths and preferences, resulting in varying levels of performance. In this paper, we compare the capabilities of the most advanced LLMs--DeepSeek, ChatGPT, and Claude--along with their reasoning-optimized versions in addressing computational challenges. Specifically, we evaluate their proficiency in solving traditional numerical problems in scientific computing as well as leveraging scientific machine learning techniques for PDE-based problems. We designed all our experiments so that a non-trivial decision is required, e.g. defining the proper space of input functions for neural operator learning. Our findings show that reasoning and hybrid-reasoning models consistently and significantly outperform non-reasoning ones in solving challenging problems, with ChatGPT o3-mini-high generally offering the fastest reasoning speed.

cs.LG↗

arXiv2025

Scalable Bayesian Physics-Informed Kolmogorov-Arnold Networks

Uncertainty quantification (UQ) plays a pivotal role in scientific machine learning, especially when surrogate models are used to approximate complex systems. Although multilayer perceptions (MLPs) are commonly employed as surrogates, they often suffer from overfitting due to their large number of parameters. Kolmogorov-Arnold networks (KANs) offer an alternative solution with fewer parameters. However, gradient-based inference methods, such as Hamiltonian Monte Carlo (HMC), may result in computational inefficiency when applied to KANs, especially for large-scale datasets, due to the high cost of back-propagation. To address these challenges, we propose a novel approach, combining the dropout Tikhonov ensemble Kalman inversion (DTEKI) with Chebyshev KANs. This gradient-free method effectively mitigates overfitting and enhances numerical stability. Additionally, we incorporate the active subspace method to reduce the parameter-space dimensionality, allowing us to improve the accuracy of predictions and obtain more reliable uncertainty estimates. Extensive experiments demonstrate the efficacy of our approach in various test cases, including scenarios with large datasets and high noise levels. Our results show that the new method achieves comparable or better accuracy, much higher efficiency as well as stability compared to HMC, in addition to scalability. Moreover, by leveraging the low-dimensional parameter subspace, our method preserves prediction accuracy while substantially reducing further the computational cost.

math.NA↗

arXiv2024

Adaptive operator learning for infinite-dimensional Bayesian inverse problems

The fundamental computational issues in Bayesian inverse problems (BIP) governed by partial differential equations (PDEs) stem from the requirement of repeated forward model evaluations. A popular strategy to reduce such costs is to replace expensive model simulations with computationally efficient approximations using operator learning, motivated by recent progress in deep learning. However, using the approximated model directly may introduce a modeling error, exacerbating the already ill-posedness of inverse problems. Thus, balancing between accuracy and efficiency is essential for the effective implementation of such approaches. To this end, we develop an adaptive operator learning framework that can reduce modeling error gradually by forcing the surrogate to be accurate in local areas. This is accomplished by adaptively fine-tuning the pre-trained approximate model with training points chosen by a greedy algorithm during the posterior evaluation process. To validate our approach, we use DeepOnet to construct the surrogate and unscented Kalman inversion (UKI) to approximate the BIP solution, respectively. Furthermore, we present a rigorous convergence guarantee in the linear case using the UKI framework. The approach is tested on a number of benchmarks, including the Darcy flow, the heat source inversion problem, and the reaction-diffusion problem. The numerical results show that our method can significantly reduce computational costs while maintaining inversion accuracy.

math.NA↗

arXiv2023

Failure-informed adaptive sampling for PINNs, Part II: combining with re-sampling and subset simulation

This is the second part of our series works on failure-informed adaptive sampling for physic-informed neural networks (FI-PINNs). In our previous work \cite{gao2022failure}, we have presented an adaptive sampling framework by using the failure probability as the posterior error indicator, where the truncated Gaussian model has been adopted for estimating the indicator. In this work, we present two novel extensions to FI-PINNs. The first extension consist in combining with a re-sampling technique, so that the new algorithm can maintain a constant training size. This is achieved through a cosine-annealing, which gradually transforms the sampling of collocation points from uniform to adaptive via training progress. The second extension is to present the subset simulation algorithm as the posterior model (instead of the truncated Gaussian model) for estimating the error indicator, which can more effectively estimate the failure probability and generate new effective training points in the failure region. We investigate the performance of the new approach using several challenging problems, and numerical experiments demonstrate a significant improvement over the original algorithm.

math.NA↗

arXiv2023

Failure-informed adaptive sampling for PINNs

Physics-informed neural networks (PINNs) have emerged as an effective technique for solving PDEs in a wide range of domains. It is noticed, however, the performance of PINNs can vary dramatically with different sampling procedures. For instance, a fixed set of (prior chosen) training points may fail to capture the effective solution region (especially for problems with singularities). To overcome this issue, we present in this work an adaptive strategy, termed the failure-informed PINNs (FI-PINNs), which is inspired by the viewpoint of reliability analysis. The key idea is to define an effective failure probability based on the residual, and then, with the aim of placing more samples in the failure region, the FI-PINNs employs a failure-informed enrichment technique to adaptively add new collocation points to the training set, such that the numerical accuracy is dramatically improved. In short, similar as adaptive finite element methods, the proposed FI-PINNs adopts the failure probability as the posterior error indicator to generate new training points. We prove rigorous error bounds of FI-PINNs and illustrate its performance through several problems.

math.NA↗

arXiv2022

A Sufficient Condition for Convex Hull Property in General Convex Spatio-Temporal Corridors

Motion planning is one of the key modules in autonomous driving systems to generate trajectories for self-driving vehicles to follow. A common motion planning approach is to generate trajectories within semantic safe corridors. The trajectories are generated by optimizing parametric curves (e.g. Bezier curves) according to an objective function. To guarantee safety, the curves are required to satisfy the convex hull property, and be contained within the safety corridors. The convex hull property however does not necessary hold for time-dependent corridors, and depends on the shape of corridors. The existing approaches only support simple shape corridors, which is restrictive in real-world, complex scenarios. In this paper, we provide a sufficient condition for general convex, spatio-temporal corridors with theoretical proof of guaranteed convex hull property. The theorem allows for using more complicated shapes to generate spatio-temporal corridors and minimizing the uncovered search space to $O(\frac{1}{n^2})$ compared to $O(1)$ of trapezoidal corridors, which can improve the optimality of the solution. Simulation results show that using general convex corridors yields less harsh brakes, hence improving the overall smoothness of the resulting trajectories.

cs.RO↗

arXiv2021

A Variational Integrator for the Discrete Element Method

A novel implicit integration scheme for the Discrete Element Method (DEM) based on the variational integrator approach is presented. The numerical solver provides a fully dynamical description that, notably, reduces to an energy minimisation scheme in the quasi-static limit. A detailed derivation of the numerical method is presented for the Hookean contact model and tested against an established open source DEM package that uses the velocity-Verlet integration scheme. These tests compare results for a single collision, long-term stability and statistical quantities of ensembles of particles. Numerically, the proposed integration method demonstrates equivalent accuracy to the velocity-Verlet method.

math.NA↗

arXiv2020

Modelling and Synchronisation of Delayed Packet-Coupled Oscillators in Industrial Wireless Sensor Networks

In this paper, a Packet-Coupled Oscillators (PkCOs) synchronisation protocol is proposed for time-sensitive Wireless Sensor Networks (WSNs) based on Pulse-Coupled Oscillators (PCO) in mathematical biology. The effects of delays on synchronisation performance are studied through mathematical modelling and analysis of packet exchange and processing delays. The delay compensation strategy (i.e., feedforward control) is utilised to cancel delays effectively. A simple scheduling function is provided with PkCOs to allocate the packet transmission event to a specified time slot, by configuring reference input of the system to a non-zero value, in order to minimise the possibility of packet collision in synchronised wireless networks. The rigorous theoretical proofs are provided to validate the convergence and stability of the proposed synchronisation scheme. Finally, the simulations and experiments examine the effectiveness of PkCOs with delay compensation and scheduling strategies. The experimental results also show that the proposed PkCOs algorithm can achieve synchronisation with the precision of $26.3μs$ ($1$ tick).

eess.SY↗

arXiv2020

NAIST COVID: Multilingual COVID-19 Twitter and Weibo Dataset

Since the outbreak of coronavirus disease 2019 (COVID-19) in the late 2019, it has affected over 200 countries and billions of people worldwide. This has affected the social life of people owing to enforcements, such as "social distancing" and "stay at home." This has resulted in an increasing interaction through social media. Given that social media can bring us valuable information about COVID-19 at a global scale, it is important to share the data and encourage social media studies against COVID-19 or other infectious diseases. Therefore, we have released a multilingual dataset of social media posts related to COVID-19, consisting of microblogs in English and Japanese from Twitter and those in Chinese from Weibo. The data cover microblogs from January 20, 2020, to March 24, 2020. This paper also provides a quantitative as well as qualitative analysis of these datasets by creating daily word clouds as an example of text-mining analysis. The dataset is now available on Github. This dataset can be analyzed in a multitude of ways and is expected to help in efficient communication of precautions related to COVID-19.

cs.SI↗

arXiv2016

Resonant TMR inversion in LiF/EuS based spin-filter tunnel junctions

Resonant tunneling can lead to inverse tunnel magnetoresistance when impurity levels rather than direct tunneling dominate the transport process. We fabricated hybrid magnetic tunnel junctions of CoFe/LiF/EuS/Ti, with an epitaxial LiF energy barrier joined with a polycrystalline EuS spin-filter bar-rier. Due to the water solubility of LiF, the devices were fully packaged in situ. The devices showed sizeable positive TMR up to 16% at low bias voltages but clearly inverted TMR at higher bias voltages. The TMR inversion depends sensitively on the thickness of LiF, and the tendency of inversion disap-pears when LiF gets thick enough and recovers its intrinsic properties.

cond-mat.mtrl-sci↗