arXiv subjects

Han Wang

Publications and source records attributed to Han Wang.

At least 217 records · Page 12Linked to original sources

arXiv2025

Data-Enabled Predictive Control for Nonlinear Systems Based on a Koopman Bilinear Realization

This paper extends the Willems' Fundamental Lemma to nonlinear control-affine systems using the Koopman bilinear realization. This enables us to bypass the Extended Dynamic Mode Decomposition (EDMD)-based system identification step in conventional Koopman-based methods and design controllers for nonlinear systems directly from data. Leveraging this result, we develop a Data-Enabled Predictive Control (DeePC) framework for nonlinear systems with unknown dynamics. A case study demonstrates that our direct data-driven control method achieves improved optimality compared to conventional Koopman-based methods. Furthermore, in examples where an exact Koopman realization with a finite-dimensional lifting function set of the controlled nonlinear system does not exist, our method exhibits advanced robustness to finite Koopman approximation errors compared to existing methods.

math.OC↗

arXiv2025

Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks

Vision Language Models (VLMs) can produce unintended and harmful content when exposed to adversarial attacks, particularly because their vision capabilities create new vulnerabilities. Existing defenses, such as input preprocessing, adversarial training, and response evaluation-based methods, are often impractical for real-world deployment due to their high costs. To address this challenge, we propose ASTRA, an efficient and effective defense by adaptively steering models away from adversarial feature directions to resist VLM attacks. Our key procedures involve finding transferable steering vectors representing the direction of harmful response and applying adaptive activation steering to remove these directions at inference time. To create effective steering vectors, we randomly ablate the visual tokens from the adversarial images and identify those most strongly associated with jailbreaks. These tokens are then used to construct steering vectors. During inference, we perform the adaptive steering method that involves the projection between the steering vectors and calibrated activation, resulting in little performance drops on benign inputs while strongly avoiding harmful outputs under adversarial inputs. Extensive experiments across multiple models and baselines demonstrate our state-of-the-art performance and high efficiency in mitigating jailbreak risks. Additionally, ASTRA exhibits good transferability, defending against unseen attacks (i.e., structured-based attack, perturbation-based attack with project gradient descent variants, and text-only attack). Our code is available at \url{https://github.com/ASTRAL-Group/ASTRA}.

cs.CV↗

arXiv2025

Fine-Tuning without Performance Degradation

Fine-tuning policies learned offline remains a major challenge in application domains. Monotonic performance improvement during \emph{fine-tuning} is often challenging, as agents typically experience performance degradation at the early fine-tuning stage. The community has identified multiple difficulties in fine-tuning a learned network online, however, the majority of progress has focused on improving learning efficiency during fine-tuning. In practice, this comes at a serious cost during fine-tuning: initially, agent performance degrades as the agent explores and effectively overrides the policy learned offline. We show across a range of settings, many offline-to-online algorithms exhibit either (1) performance degradation or (2) slow learning (sometimes effectively no improvement) during fine-tuning. We introduce a new fine-tuning algorithm, based on an algorithm called Jump Start, that gradually allows more exploration based on online estimates of performance. Empirically, this approach achieves fast fine-tuning and significantly reduces performance degradations compared with existing algorithms designed to do the same.

cs.LG↗

arXiv2025

PRISM: Projection-based Reward Integration for Scene-Aware Real-to-Sim-to-Real Transfer with Few Demonstrations

Learning from few demonstrations to develop policies robust to variations in robot initial positions and object poses is a problem of significant practical interest in robotics. Compared to imitation learning, which often struggles to generalize from limited samples, reinforcement learning (RL) can autonomously explore to obtain robust behaviors. Training RL agents through direct interaction with the real world is often impractical and unsafe, while building simulation environments requires extensive manual effort, such as designing scenes and crafting task-specific reward functions. To address these challenges, we propose an integrated real-to-sim-to-real pipeline that constructs simulation environments based on expert demonstrations by identifying scene objects from images and retrieving their corresponding 3D models from existing libraries. We introduce a projection-based reward model for RL policy training that is supervised by a vision-language model (VLM) using human-guided object projection relationships as prompts, with the policy further fine-tuned using expert demonstrations. In general, our work focuses on the construction of simulation environments and RL-based policy training, ultimately enabling the deployment of reliable robotic control policies in real-world scenarios.

cs.RO↗

arXiv2025

AdaCAD: Adaptively Decoding to Balance Conflicts between Contextual and Parametric Knowledge

Knowledge conflict arises from discrepancies between information in the context of a large language model (LLM) and the knowledge stored in its parameters. This can hurt performance when using standard decoding techniques, which tend to ignore the context. Existing test-time contrastive methods seek to address this by comparing the LLM's output distribution with and without the context and adjust the model according to the contrast between them. However, we find that these methods frequently misjudge the degree of conflict and struggle to handle instances that vary in their amount of conflict, with static methods over-adjusting when conflict is absent. We propose a fine-grained, instance-level approach called AdaCAD, which dynamically infers the weight of adjustment based on the degree of conflict, as measured by the Jensen-Shannon divergence between distributions representing contextual and parametric knowledge. Across four LLMs, six question-answering (QA) and three summarization datasets, we demonstrate that ADACAD consistently outperforms other decoding baselines with average QA accuracy gains of 14.21% (absolute) over a static contrastive baseline, and improves the factuality of summaries by 6.19 (AlignScore). Lastly, we show that while contrastive baselines hurt performance when conflict is absent, ADACAD mitigates these losses, making it more applicable to real-world datasets in which some examples have conflict and others do not.

cs.CL↗

arXiv2025

Opt-ODENet: A Neural ODE Framework with Differentiable QP Layers for Safe and Stable Control Design (longer version)

Designing controllers that achieve task objectives while ensuring safety is a key challenge in control systems. This work introduces Opt-ODENet, a Neural ODE framework with a differentiable Quadratic Programming (QP) optimization layer to enforce constraints as hard requirements. Eliminating the reliance on nominal controllers or large datasets, our framework solves the optimal control problem directly using Neural ODEs. Stability and convergence are ensured through Control Lyapunov Functions (CLFs) in the loss function, while Control Barrier Functions (CBFs) embedded in the QP layer enforce real-time safety. By integrating the differentiable QP layer with Neural ODEs, we demonstrate compatibility with the adjoint method for gradient computation, enabling the learning of the CBF class-$\mathcal{K}$ function and control network parameters. Experiments validate its effectiveness in balancing safety and performance.

eess.SY↗

arXiv2025

COUNTER: Cluster GCN based Energy Efficient Resource Management for Sustainable Cloud Computing Environments

Cloud computing, thanks to the pervasiveness of information technologies, provides a foundational environment for developing IT applications, offering organizations virtually unlimited and flexible computing resources on a pay-per-use basis. However, the large data centres where cloud computing services are hosted consume significant amounts of electricity annually due to Information and Communication Technology (ICT) components. This issue is exacerbated by the increasing deployment of large artificial intelligence (AI) models, which often rely on distributed data centres, thereby significantly impacting the global environment. This study proposes the COUNTER model, designed for sustainable cloud resource management. COUNTER is integrated with cluster graph neural networks and evaluated in a simulated cloud environment, aiming to reduce energy consumption while maintaining quality of service parameters. Experimental results demonstrate improvements in resource utilisation, energy consumption, and cost effectiveness compared to the baseline model, HUNTER, which employs a gated graph neural network aimed at achieving carbon neutrality in cloud computing for modern ICT systems.

cs.DC↗

arXiv2025

Innovative Thinking, Infinite Humor: Humor Research of Large Language Models through Structured Thought Leaps

Humor is previously regarded as a gift exclusive to humans for the following reasons. Humor is a culturally nuanced aspect of human language, presenting challenges for its understanding and generation. Humor generation necessitates a multi-hop reasoning process, with each hop founded on proper rationales. Although many studies, such as those related to GPT-o1, focus on logical reasoning with reflection and correction, they still fall short in humor generation. Due to the sparsity of the knowledge graph in creative thinking, it is arduous to achieve multi-hop reasoning. Consequently, in this paper, we propose a more robust framework for addressing the humor reasoning task, named LoL. LoL aims to inject external information to mitigate the sparsity of the knowledge graph, thereby enabling multi-hop reasoning. In the first stage of LoL, we put forward an automatic instruction-evolution method to incorporate the deeper and broader thinking processes underlying humor. Judgment-oriented instructions are devised to enhance the model's judgment capability, dynamically supplementing and updating the sparse knowledge graph. Subsequently, through reinforcement learning, the reasoning logic for each online-generated response is extracted using GPT-4o. In this process, external knowledge is re-introduced to aid the model in logical reasoning and the learning of human preferences. Finally, experimental results indicate that the combination of these two processes can enhance both the model's judgment ability and its generative capacity. These findings deepen our comprehension of the creative capabilities of large language models (LLMs) and offer approaches to boost LLMs' creative abilities for cross-domain innovative applications.

cs.AI↗

arXiv2025

Nonlinear Optimal Guidance for Intercepting Moving Targets

This paper introduces a nonlinear optimal guidance framework for guiding a pursuer to intercept a moving target, with an emphasis on real-time generation of optimal feedback control for a nonlinear optimal control problem. Initially, considering the target moves without maneuvering, we derive the necessary optimality conditions using Pontryagin's Maximum Principle. These conditions reveal that each extremal trajectory is uniquely determined by two scalar parameters. Analyzing the geometric property of the parameterized extremal trajectories not only leads to an additional necessary condition but also allows to establish a sufficient condition for local optimality. This enables the generation of a dataset containing at least locally optimal trajectories. By studying the properties of the optimal feedback control, the size of the dataset is reduced significantly, allowing training a lightweight neural network to predict the optimal guidance command in real time. Furthermore, the performance of the neural network is enhanced by incorporating the target's acceleration, making it suitable for intercepting both uniformly moving and maneuvering targets. Finally, numerical simulations validate the proposed nonlinear optimal guidance framework, demonstrating its better performance over existing guidance laws.

math.OC↗

arXiv2025

Confidence-Aware Learning Optimal Terminal Guidance via Gaussian Process Regression

Modern aerospace guidance systems demand rigorous constraint satisfaction, optimal performance, and computational efficiency. Traditional analytical methods struggle to simultaneously satisfy these requirements. While data driven methods have shown promise in learning optimal guidance strategy, challenges still persist in generating well-distributed optimal dataset and ensuring the reliability and trustworthiness of learned strategies. This paper presents a confidence-aware learning framework that addresses these limitations. First, a region-controllable optimal data generation method is proposed leveraging Hamiltonian state transition matrices, enabling efficient generation of optimal trajectories of specified data distribution. Then, to obtain a lightweight and effective dataset for efficient strategy learning, an error-distribution-smoothing method is incorporated to employ data filtering, which reduces dataset size by almost 90% while preserving prediction accuracy. To assess the operational domain of the learned strategy, a confidence-aware learning guidance strategy is proposed based on gaussian process regression, achieving constraint satisfaction even beyond training distributions. Numerical simulations validate the effectiveness and reliability of the proposed learning framework in terms of data generation, data filtering and strategy learning.

eess.SY↗

arXiv2025

Nanomolding single-crystalline CoIn3 and RhIn3 nanowires

Intermetallic compounds containing transition metals and group III-V metals tend to possess strong correlations and high catalytic activities, both of which can be enhanced via reduced dimensionality. Nanostructuring is an effective approach to explore this possibility, yet the synthesis of nanostructured intermetallics is challenging due to vast differences in melting points and vapor pressures of the constituent elements. In this work, we demonstrate that this challenge can be overcome with thermomechanical nanomolding (TMNM), exemplified by the synthesis of intermetallic CoIn3 and RhIn3 nanowires. We show that TMNM successfully extrudes single-crystalline nanowires of these compounds down to the 20 nm diameter range, and the nanowires remain metallic with resistivity values higher than calculated bulk resistivity. We discuss possible effects of surface roughness scattering, vacancy-induced scattering, and surface oxidation, on the measured resistivities of the nanowires. For CoIn3 nanowires, the measured resistivity values are the first reported values for this compound.

cond-mat.mtrl-sci↗

arXiv2025

Vision as LoRA

We introduce Vision as LoRA (VoRA), a novel paradigm for transforming an LLM into an MLLM. Unlike prevalent MLLM architectures that rely on external vision modules for vision encoding, VoRA internalizes visual capabilities by integrating vision-specific LoRA layers directly into the LLM. This design allows the added parameters to be seamlessly merged into the LLM during inference, eliminating structural complexity and minimizing computational overhead. Moreover, inheriting the LLM's ability of handling flexible context, VoRA can process inputs at arbitrary resolutions. To further strengthen VoRA's visual capabilities, we introduce a block-wise distillation method that transfers visual priors from a pre-trained ViT into the LoRA layers, effectively accelerating training by injecting visual knowledge. Additionally, we apply bi-directional attention masks to better capture the context information of an image. We successfully demonstrate that with additional pre-training data, VoRA can perform comparably with conventional encode-based MLLMs. All training data, codes, and model weights will be released at https://github.com/Hon-Wong/VoRA.

cs.CV↗

arXiv2025

Constraint Horizon in Model Predictive Control

In this work, we propose a Model Predictive Control (MPC) formulation incorporating two distinct horizons: a prediction horizon and a constraint horizon. This approach enables a deeper understanding of how constraints influence key system properties such as suboptimality, without compromising recursive feasibility and constraint satisfaction. In this direction, our contributions are twofold. First, we provide a framework to estimate closed-loop optimality as a function of the number of enforced constraints. This is a generalization of existing results by considering partial constraint enforcement over the prediction horizon. Second, when adopting this general framework under the lens of safety-critical applications, our method improves conventional Control Barrier Function (CBF) based approaches. It mitigates myopic behaviour in Quadratic Programming (QP)-CBF schemes, and resolves compatibility issues between Control Lyapunov Function (CLF) and CBF constraints via the prediction horizon used in the optimization. We show the efficacy of the method via numerical simulations for a safety critical application.

eess.SY↗

arXiv2025

Distributed Safe Control Design and Probabilistic Safety Verification for Multi-Agent Systems

We propose distributed iterative algorithms for safe control design and safety verification for networked multi-agent systems. These algorithms rely on distributing a control barrier function (CBF) related quadratic programming (QP) problem assuming the existence of CBFs. The proposed distributed algorithm addresses infeasibility issues of existing schemes via a cooperation mechanism between agents. The resulting control input is guaranteed to be optimal, and satisfies CBF constraints of all agents. Furthermore, a truncated algorithm is proposed to facilitate computational implementation. The performance of the truncated algorithm is evaluated using a distributed safety verification algorithm. The algorithm quantifies safety for multi-agent systems probabilistically by means of CBFs. Both upper and lower bounds on the probability of safety are obtained using the so called scenario approach. Both the scenario sampling and safety verification procedures are fully distributed. The efficacy of our algorithms is demonstrated by an example on multi-robot collision avoidance.

eess.SY↗

arXiv2025

The Emperor's New Clothes in Benchmarking? A Rigorous Examination of Mitigation Strategies for LLM Benchmark Data Contamination

Benchmark Data Contamination (BDC)-the inclusion of benchmark testing samples in the training set-has raised increasing concerns in Large Language Model (LLM) evaluation, leading to falsely inflated performance estimates and undermining evaluation reliability. To address this, researchers have proposed various mitigation strategies to update existing benchmarks, including modifying original questions or generating new ones based on them. However, a rigorous examination of the effectiveness of these mitigation strategies remains lacking. In this paper, we design a systematic and controlled pipeline along with two novel metrics-fidelity and contamination resistance-to provide a fine-grained and comprehensive assessment of existing BDC mitigation strategies. Previous assessment methods, such as accuracy drop and accuracy matching, focus solely on aggregate accuracy, often leading to incomplete or misleading conclusions. Our metrics address this limitation by emphasizing question-level evaluation result matching. Extensive experiments with 10 LLMs, 5 benchmarks, 20 BDC mitigation strategies, and 2 contamination scenarios reveal that no existing strategy significantly improves resistance over the vanilla case (i.e., no benchmark update) across all benchmarks, and none effectively balances fidelity and contamination resistance. These findings underscore the urgent need for designing more effective BDC mitigation strategies. Our code repository is available at https://github.com/ASTRAL-Group/BDC_mitigation_assessment.

cs.AI↗

arXiv2025

Dynamic Open-Vocabulary 3D Scene Graphs for Long-term Language-Guided Mobile Manipulation

Enabling mobile robots to perform long-term tasks in dynamic real-world environments is a formidable challenge, especially when the environment changes frequently due to human-robot interactions or the robot's own actions. Traditional methods typically assume static scenes, which limits their applicability in the continuously changing real world. To overcome these limitations, we present DovSG, a novel mobile manipulation framework that leverages dynamic open-vocabulary 3D scene graphs and a language-guided task planning module for long-term task execution. DovSG takes RGB-D sequences as input and utilizes vision-language models (VLMs) for object detection to obtain high-level object semantic features. Based on the segmented objects, a structured 3D scene graph is generated for low-level spatial relationships. Furthermore, an efficient mechanism for locally updating the scene graph, allows the robot to adjust parts of the graph dynamically during interactions without the need for full scene reconstruction. This mechanism is particularly valuable in dynamic environments, enabling the robot to continually adapt to scene changes and effectively support the execution of long-term tasks. We validated our system in real-world environments with varying degrees of manual modifications, demonstrating its effectiveness and superior performance in long-term tasks. Our project page is available at: https://bjhyzj.github.io/dovsg-web.

cs.RO↗

arXiv2025

Surface-dominant transport in Weyl semimetal NbAs nanowires for next-generation interconnects

Ongoing demands for smaller and more energy efficient electronic devices necessitate alternative interconnect materials with lower electrical resistivity at reduced dimensions. Despite the emergence of many promising candidates, synthesizing high quality nanostructures remains a major bottleneck in evaluating their performance. Here, we report the successful synthesis of Weyl semimetal NbAs nanowires via thermomechanical nanomolding, achieving single crystallinity and controlled diameters as small as 40 nm. Our NbAs nanowires exhibit a remarkably low room-temperature resistivity of 9.7 +/- 1.6 microOhm-cm, which is three to four times lower than their bulk counterpart. Theoretical calculations corroborate the experimental observations, attributing this exceptional resistivity reduction to surface dominant conduction with long carrier lifetime at finite temperatures. Further characterization of NbAs nanowires and bulk single crystals reveals high breakdown current density, robust stability, and superior thermal conductivity. Collectively, these properties highlight the strong potential of NbAs nanowires as next-generation interconnects, which can surpass the limitations of current copper-based interconnects. Technologically, our findings present a practical application of topological materials, while scientifically showcasing the fundamental properties uniquely accessible in nanoscale platforms.

cond-mat.mes-hall↗

arXiv2025

Learning Structured Compressed Sensing with Automatic Resource Allocation

Multidimensional data acquisition often requires extensive time and poses significant challenges for hardware and software regarding data storage and processing. Rather than designing a single compression matrix as in conventional compressed sensing, structured compressed sensing yields dimension-specific compression matrices, reducing the number of optimizable parameters. Recent advances in machine learning (ML) have enabled task-based supervised learning of subsampling matrices, albeit at the expense of complex downstream models. Additionally, the sampling resource allocation across dimensions is often determined in advance through heuristics. To address these challenges, we introduce Structured COmpressed Sensing with Automatic Resource Allocation (SCOSARA) with an information theory-based unsupervised learning strategy. SCOSARA adaptively distributes samples across sampling dimensions while maximizing Fisher information content. Using ultrasound localization as a case study, we compare SCOSARA to state-of-the-art ML-based and greedy search algorithms. Simulation results demonstrate that SCOSARA can produce high-quality subsampling matrices that achieve lower Cramér-Rao Bound values than the baselines. In addition, SCOSARA outperforms other ML-based algorithms in terms of the number of trainable parameters, computational complexity, and memory requirements while automatically choosing the number of samples per axis.

cs.LG↗