arXiv subjects

Bin Han

Publications and source records attributed to Bin Han.

At least 73 records · Page 4Linked to original sources

arXiv2025

Do Language Models Mirror Human Confidence? Exploring Psychological Insights to Address Overconfidence in LLMs

Psychology research has shown that humans are poor at estimating their performance on tasks, tending towards underconfidence on easy tasks and overconfidence on difficult tasks. We examine three LLMs, Llama-3-70B-instruct, Claude-3-Sonnet, and GPT-4o, on a range of QA tasks of varying difficulty, and show that models exhibit subtle differences from human patterns of overconfidence: less sensitive to task difficulty, and when prompted to answer based on different personas -- e.g., expert vs layman, or different race, gender, and ages -- the models will respond with stereotypically biased confidence estimations even though their underlying answer accuracy remains the same. Based on these observations, we propose Answer-Free Confidence Estimation (AFCE) to improve confidence calibration and LLM interpretability in these settings. AFCE is a self-assessment method that employs two stages of prompting, first eliciting only confidence scores on questions, then asking separately for the answer. Experiments on the MMLU and GPQA datasets spanning subjects and difficulty show that this separation of tasks significantly reduces overconfidence and delivers more human-like sensitivity to task difficulty.

cs.AI↗

arXiv2025

Colored Multiset Eulerian Polynomials

Colored multiset Eulerian polynomials are a common generalization of MacMahon's multiset Eulerian polynomials and the colored Eulerian polynomials, both of which are known to satisfy well-studied distributional properties including real-rootedness, log-concavity and unimodality. The symmetric colored multiset Eulerian polynomials are characterized and used to prove sufficient conditions for a colored multiset Eulerian polynomial to be self-interlacing. The latter property implies the aforementioned distributional properties as well as others, including the alternatingly increasing property and bi-$γ$-positivity. To derive these results, multivariate generalizations of an identity due to MacMahon are deduced. The results are applied to a pair of questions, both previously studied in several special cases, that are seen to admit more general answers when framed in the context of colored multiset Eulerian polynomials. The first question pertains to $s$-Eulerian polynomials, and the second to interpretations of $γ$-coefficients.

math.CO↗

arXiv2025

Salience Adjustment for Context-Based Emotion Recognition

Emotion recognition in dynamic social contexts requires an understanding of the complex interaction between facial expressions and situational cues. This paper presents a salience-adjusted framework for context-aware emotion recognition with Bayesian Cue Integration (BCI) and Visual-Language Models (VLMs) to dynamically weight facial and contextual information based on the expressivity of facial cues. We evaluate this approach using human annotations and automatic emotion recognition systems in prisoner's dilemma scenarios, which are designed to evoke emotional reactions. Our findings demonstrate that incorporating salience adjustment enhances emotion recognition performance, offering promising directions for future research to extend this framework to broader social contexts and multimodal applications.

cs.CV↗

arXiv2025

Vibration-aware Lidar-Inertial Odometry based on Point-wise Post-Undistortion Uncertainty

High-speed ground robots moving on unstructured terrains generate intense high-frequency vibrations, leading to LiDAR scan distortions in Lidar-inertial odometry (LIO). Accurate and efficient undistortion is extremely challenging due to (1) rapid and non-smooth state changes during intense vibrations and (2) unpredictable IMU noise coupled with a limited IMU sampling frequency. To address this issue, this paper introduces post-undistortion uncertainty. First, we model the undistortion errors caused by linear and angular vibrations and assign post-undistortion uncertainty to each point. We then leverage this uncertainty to guide point-to-map matching, compute uncertainty-aware residuals, and update the odometry states using an iterated Kalman filter. We conduct vibration-platform and mobile-platform experiments on multiple public datasets as well as our own recordings, demonstrating that our method achieves better performance than other methods when LiDAR undergoes intense vibration.

cs.RO↗

arXiv2025

Hardware-Free Event Cameras Temporal Synchronization Based on Event Density Alignment

Event cameras are a novel type of sensor designed for capturing the dynamic changes of a scene. Due to factors such as trigger and transmission delays, a time offset exists in the data collected by multiple event cameras, leading to inaccurate information fusion. Thus, the collected data needs to be synchronized to overcome any potential time offset issue. Hardware synchronization methods require additional circuits, while certain models of event cameras (e.g., CeleX5) do not support hardware synchronization. Therefore, this paper proposes a hardware-free event camera synchronization method. This method determines differences between start times by minimizing the dissimilarity of the event density distributions of different event cameras and synchronizes the data by adjusting timestamps. The experiments demonstrate that the method's synchronization error is less than 10ms under various senses with multiple models of event cameras.

cs.RO↗

arXiv2025

Convergent Sixth-order Compact Finite Difference Method for Variable-Coefficient Elliptic PDEs in Curved Domains

Finite difference methods (FDMs) are widely used for solving partial differential equations (PDEs) due to their relatively simple implementation. However, they face significant challenges when applied to non-rectangular domains and in establishing theoretical convergence, particularly for high-order schemes. In this paper, we focus on solving the elliptic equation $-\nabla \cdot (a \nabla u) = f$ in a two-dimensional curved domain $Ω$, where the diffusion coefficient $a$ is variable and smooth. We propose a sixth-order $9$-point compact FDM on uniform Cartesian grids within the domain, not relying on ghost points or information outside $\overlineΩ$. All the boundary stencils near $\partial Ω$ have at most $6$ different configurations and use at most $8$ grid points inside $Ω$. We rigorously establish the sixth-order convergence of the numerically approximated solution $u_h$ in the $\infty$-norm. Additionally, we derive a gradient approximation $\nabla u$ directly from $u_h$ without solving auxiliary equations. This gradient approximation achieves proven accuracy of order $5 + \frac{1}{q}$ in the $q$-norm for all $1 \leq q \leq \infty$ (with a logarithmic factor $\log h$ for $1 \leq q < 2$). To validate our proposed sixth-order compact finite different method, we provide several numerical examples that illustrate the sixth-order accuracy and computational efficiency of both the numerical solution and the gradient approximation for solving elliptic PDEs in curved domains.

math.NA↗

arXiv2025

Lightweight Node Selection in Hexagonal Grid Topology for TDoA-Based UAV Localization

This paper investigates the optimization problem for TDoA-based UAV localization in low-altitude urban environments with hexagonal grid node deployment. We derive a lightweight optimized node selection strategy based on only RSSI measurements, to pre-select optimal nodes, avoiding extensive TDoA measurements in energy-constrained UAV scenarios. Theoretical and simulation results demonstrate that dynamically selecting the number of reference nodes improves localization performance while minimizing resource overhead.

eess.SP↗

arXiv2025

A High-Order, Pressure-Robust, and Decoupled Finite Difference Method for the Stokes Problem

In this paper, we consider the Stokes problem with Dirichlet boundary conditions and the constant kinematic viscosity $ν$ in an axis-aligned domain $Ω$. We decouple the velocity $\bm u$ and pressure $p$ by deriving a novel biharmonic equation in $Ω$ and third-order boundary conditions on $\partialΩ$. In contrast to the fourth-order streamfunction approach, our formulation does not require $Ω$ to be simply connected. For smooth velocity fields $\bm u$ in two dimensions, we explicitly construct a finite difference method (FDM) with sixth-order consistency to approximate $\bm u$ at all relevant grid points: interior points, boundary side points, and boundary corner points. The resulting scheme yields two linear systems $A_1u^{(1)}_h=b_1$ and $A_2u^{(2)}_h=b_2$, where $A_1,A_2$ are constant matrices, and $b_1,b_2$ are independent of the pressure $p$ and the kinematic viscosity $ν$. Thus, the proposed method is pressure- and viscosity-robust. To accommodate velocity fields with less regularity, we modify the FDM by removing singular terms in the right-hand side vectors. Once the discrete velocity is computed, we apply a sixth-order finite difference operator to approximate the pressure gradient locally, without solving any additional linear systems. In our numerical experiments, we test both smooth and non-smooth solutions $(\bm u,p)$ in a square domain, a triply connected domain, and an $L$-shaped domain in two dimensions. The results confirm sixth-order convergence of the velocity and pressure gradient in the $\ell_\infty$-norm for smooth solutions. For non-smooth velocity fields, our method achieves the expected lower-order convergence. Moreover, the observed velocity error $\|{\bm u}_h-\bm u\|_{\infty}$ is independent of the pressure $p$ and viscosity $ν$.

math.NA↗

arXiv2025

Why Not Replace? Sustaining Long-Term Visual Localization via Handcrafted-Learned Feature Collaboration on CPU

Robust long-term visual localization in complex industrial environments is critical for mobile robotic systems. Existing approaches face limitations: handcrafted features are illumination-sensitive, learned features are computationally intensive, and semantic- or marker-based methods are environmentally constrained. Handcrafted and learned features share similar representations but differ functionally. Handcrafted features are optimized for continuous tracking, while learned features excel in wide-baseline matching. Their complementarity calls for integration rather than replacement. Building on this, we propose a hierarchical localization framework. It leverages real-time handcrafted feature extraction for relative pose estimation. In parallel, it employs selective learned keypoint detection on optimized keyframes for absolute positioning. This design enables CPU-efficient, long-term visual localization. Experiments systematically progress through three validation phases: Initially establishing feature complementarity through comparative analysis, followed by computational latency profiling across algorithm stages on CPU platforms. Final evaluation under photometric variations (including seasonal transitions and diurnal cycles) demonstrates 47% average error reduction with significantly improved localization consistency. The code implementation is publicly available at https://github.com/linyicheng1/ORB_SLAM3_localization.

cs.CV↗

arXiv2025

Fragments to Facts: Partial-Information Fragment Inference from LLMs

Large language models (LLMs) can leak sensitive training data through memorization and membership inference attacks. Prior work has primarily focused on strong adversarial assumptions, including attacker access to entire samples or long, ordered prefixes, leaving open the question of how vulnerable LLMs are when adversaries have only partial, unordered sample information. For example, if an attacker knows a patient has "hypertension," under what conditions can they query a model fine-tuned on patient data to learn the patient also has "osteoarthritis?" In this paper, we introduce a more general threat model under this weaker assumption and show that fine-tuned LLMs are susceptible to these fragment-specific extraction attacks. To systematically investigate these attacks, we propose two data-blind methods: (1) a likelihood ratio attack inspired by methods from membership inference, and (2) a novel approach, PRISM, which regularizes the ratio by leveraging an external prior. Using examples from both medical and legal settings, we show that both methods are competitive with a data-aware baseline classifier that assumes access to labeled in-distribution data, underscoring their robustness.

cs.LG↗

arXiv2025

Trustworthy UAV Cooperative Localization: Information Analysis of Performance and Security

This paper presents a trustworthy framework for achieving accurate cooperative localization in multiple unmanned aerial vehicle (UAV) systems. The The Cramer-Rao Lower Bound (CRLB) for the three-dimensional (3D) cooperative localization network is derived, with particular attention given to practical scenarios involving non-uniform spatial distribution of anchor nodes. Challenges of mobility are then addressed with Mobility Adaptive Gradient Descent (MAGD). In the context of system security, we derive the CRLB of localization under the influence of falsified information. The methods and strategies of injecting such information and their impact on system performance are studied. To assure robust performance under falsified data, we propose a mitigation solution named Time-evolving Anomaly Detection (TAD). Furthermore, we model the system performance regarding the density and magnitude of falsified information, focusing on realistic scenarios where the adversary is resource-constrained. With the vulnerability of cooperative localization understood, we apply TAD and formulate an optimization problem from the adversary's perspective. Next, we discuss the design principles of an anomaly detector, with emphasis of the trade-off of reducing such optimum and system performance. Additionally, we also deploy a reputation propagation (RP) mechanism to fully utilize the anomaly detection and further optimize the TAD. Our proposed approaches are demonstrated through numerical simulations.

eess.SP↗

arXiv2025

Wavelet Galerkin Method for an Electromagnetic Scattering Problem

The Helmholtz equation with variable wavenumbers is challenging to solve numerically due to the pollution effect, which often results in a huge ill-conditioned linear system. In this paper, we present a high-order wavelet Galerkin method to numerically solve an electromagnetic scattering from a large cavity problem modeled by the 2D Helmholtz equation with variable wavenumbers. The high approximation order and the sparse linear system with uniformly bounded condition numbers offered by wavelets are useful in dealing with the pollution effect. Using the direct approach in [B. Han and M. Michelle, Appl. Comp. Harmon. Anal., 53 (2021), 270-331], we present various optimized spline biorthogonal wavelets on a bounded interval. We provide a self-contained proof to show that the tensor product of such wavelets forms a 2D Riesz wavelet in the appropriate Sobolev space. Compared to the coefficient matrix of the finite element method (FEM), when an iterative scheme is applied to the coefficient matrix of our wavelet Galerkin method, much fewer iterations are needed for the relative residuals to be within a tolerance level. Furthermore, for a given bounded variable wavenumber, the number of required iterations is practically independent of the size of the wavelet coefficient matrix, due to the small uniformly bounded condition numbers of such wavelets. In contrast, when an iterative scheme is applied to the FEM coefficient matrix, the number of required iterations doubles as the mesh size for each axis is halved. The implementation can also be done conveniently thanks to the simple structure, the refinability property, and the analytic expression of our wavelet bases.

math.NA↗

arXiv2025

Unauthorized Radio Sensing and Privacy Risks: A Sampling Error-Based Defense

Unauthorized sensing activities pose an increasing threat to individual privacy, yet effective countermeasures remain underdeveloped. This paper presents a novel methodology to characterize and counter such unauthorized surveillance. We model pedestrian trajectories as a random process and leverage the Cramer-Rao bound (CRB) to evaluate sensing performance, interpreting it as sampling error within this random process. Through simulation, we verify our method's accuracy in monitoring unauthorized sensing activities in urban environments and validate the effectiveness of our proposed mitigation strategies.

eess.SP↗

arXiv2025

Physical Layer Deception in OFDM Systems

As a promising technology, physical layer security (PLS) enhances security by leveraging the physical characteristics of communication channels. However, it commonly takes the legitimate user more effort to secure its data, compared to that required by the eavesdropper to intercept the communication. To address this imbalance, we propose a physical layer deception (PLD) framework, which applies random deceptive ciphering combined with orthogonal frequency-division multiplexing (OFDM) to deceive eavesdroppers with falsified information, preventing them from wiretapping. While ensuring the same level of confidentiality as traditional PLS methods, the PLD approach additionally introduces a deception mechanism, which remains effective even when the eavesdropper has the same knowledge about the transmitter as the legitimate receiver. Through detailed theoretical analysis and numerical simulations, we prove the superiority of our method over the conventional PLS approach.

cs.IT↗

arXiv2025

Physical Layer Deception with Non-Orthogonal Multiplexing

Physical layer security (PLS) is a promising technology to secure wireless communications by exploiting the physical properties of the wireless channel. However, the passive nature of PLS creates a significant imbalance between the effort required by eavesdroppers and legitimate users to secure data. To address this imbalance, in this article, we propose a novel framework of physical layer deception (PLD), which combines PLS with deception technologies to actively counteract wiretapping attempts. Combining a two-stage encoder with randomized ciphering and non-orthogonal multiplexing, the PLD approach enables the wireless communication system to proactively counter eavesdroppers with deceptive messages. Relying solely on the superiority of the legitimate channel over the eavesdropping channel, the PLD framework can effectively protect the confidentiality of the transmitted messages, even against eavesdroppers who possess knowledge equivalent to that of the legitimate receiver. We prove the validity of the PLD framework with in-depth analyses and demonstrate its superiority over conventional PLS approaches with comprehensive numerical benchmarks.

cs.CR↗

arXiv2024

Quantum Machine Learning in Log-based Anomaly Detection: Challenges and Opportunities

Log-based anomaly detection (LogAD) is the main component of Artificial Intelligence for IT Operations (AIOps), which can detect anomalous that occur during the system on-the-fly. Existing methods commonly extract log sequence features using classical machine learning techniques to identify whether a new sequence is an anomaly or not. However, these classical approaches often require trade-offs between efficiency and accuracy. The advent of quantum machine learning (QML) offers a promising alternative. By transforming parts of classical machine learning computations into parameterized quantum circuits (PQCs), QML can significantly reduce the number of trainable parameters while maintaining accuracy comparable to classical counterparts. In this work, we introduce a unified framework, \ourframework{}, for evaluating QML models in the context of LogAD. This framework incorporates diverse log data, integrated QML models, and comprehensive evaluation metrics. State-of-the-art methods such as DeepLog, LogAnomaly, and LogRobust, along with their quantum-transformed counterparts, are included in our framework.Beyond standard metrics like F1 score, precision, and recall, our evaluation extends to factors critical to QML performance, such as specificity, the number of circuits, circuit design, and quantum state encoding. Using \ourframework{}, we conduct extensive experiments to assess the performance of these models and their quantum counterparts, uncovering valuable insights and paving the way for future research in QML model selection and design for LogAD.

cs.LG↗

arXiv2024

Fairness-aware Age-of-Information Minimization in WPT-Assisted Short-Packet Data Collection for mURLLC

The technological landscape is rapidly evolving toward large-scale systems. Networks supporting massive connectivity through numerous Internet of Things (IoT) devices are at the forefront of this advancement. In this paper, we examine Wireless Power Transfer (WPT)-enabled networks, where a server requires to collect data from these IoT devices to compute a task with massive Ultra-Reliable and Low-Latency Communication (mURLLC) services.} We focus on information freshness, using Age-of-Information (AoI) as the key performance metric. Specifically, we aim to minimize the maximum AoI among IoT devices by optimizing the scheduling policy. Our analytical findings demonstrate the convexity of the problem, enabling efficient solutions. We introduce the concept of AoI-oriented cluster capacity and analyze the relationship between the number of supported devices and network AoI performance. Numerical simulations validate our proposed approach's effectiveness in enhancing AoI performance, highlighting its potential for guiding the design of future IoT systems requiring mURLLC services.

cs.IT↗

arXiv2024

A Derivative-Orthogonal Wavelet Multiscale Method for 1D Elliptic Equations with Rough Diffusion Coefficients

In this paper, we investigate 1D elliptic equations $-\nabla\cdot (a\nabla u)=f$ with rough diffusion coefficients $a$ that satisfy $0<a_{\min}\le a\le a_{\max}<\infty$ and $f\in L_2(Ω)$. To achieve an accurate and robust numerical solution on a coarse mesh of size $H$, we introduce a derivative-orthogonal wavelet-based framework. This approach incorporates both regular and specialized basis functions constructed through a novel technique, defining a basis function space that enables effective approximation. We develop a derivative-orthogonal wavelet multiscale method tailored for this framework, proving that the condition number $κ$ of the stiffness matrix satisfies $κ\le a_{\max}/a_{\min}$, independent of $H$. For the error analysis, we establish that the energy and $L_2$-norm errors of our method converge at first-order and second-order rates, respectively, for any coarse mesh $H$. Specifically, the energy and $L_2$-norm errors are bounded by $2 a_{\min}^{-1/2} \|f\|_{L_2(Ω)} H$ and $4 a_{\min}^{-1}\|f\|_{L_2(Ω)} H^2$. Moreover, the numerical approximated solution also possesses the interpolation property at all grid points. We present a range of challenging test cases with continuous, discontinuous, high-frequency, and high-contrast coefficients $a$ to evaluate errors in $u, u'$ and $a u'$ in both $l_2$ and $l_\infty$ norms. We also provide a numerical example that both coefficient $a$ and source term $f$ contain discontinuous, high-frequency and high-contrast oscillations. Additionally, we compare our method with the standard second-order finite element method to assess error behaviors and condition numbers when the mesh is not fine enough to resolve coefficient oscillations. Numerical results confirm the bounded condition numbers and convergence rates, affirming the effectiveness of our approach.

math.NA↗