arXiv subjects

Yirui Zhang

Publications and source records attributed to Yirui Zhang.

15 recordsLinked to original sources

arXiv2026

SPARROW: Scalable Taxonomy Induction via Structure-Preserving Partitioning and Constraint-Guided Merging

Taxonomy induction aims to organize concept sets into coherent hierarchical structures. Recent LLM-based methods can induce taxonomies directly from flat term lists, avoiding the need for corpora, but degrade sharply as concept sets scale up. We argue that this degradation stems not only from context length limitations, but also from structural failures in hierarchical reasoning. To address this, we adopt a divide-and-merge paradigm that partitions concepts into smaller subsets, induces local taxonomies, and merges them into a global hierarchy. However, we identify two structural failure modes inherent to this paradigm: Structural Fragmentation, where partitioning weakens local hierarchical signals, and Parent Displacement, where locally plausible relations are misplaced in the global hierarchy. To address both, we propose SPARROW, a scalable taxonomy induction framework that combines structure-preserving spectral partitioning to retain hierarchical connectivity within each block, and constraint-guided incremental fusion that treats block-level relations as structural constraints rather than ground truth for global placement. Experiments on large-scale benchmarks show that SPARROW consistently achieves the strongest global structural quality across backbones. The code is available at https://github.com/rebeccazyr/SPARROW.

cs.CL↗

arXiv2026

Debating the Unspoken: Role-Anchored Multi-Agent Reasoning for Half-Truth Detection

Half-truths, claims that are factually correct yet misleading due to omitted context, remain a blind spot for fact verification systems focused on explicit falsehoods. Addressing such omission-based manipulation requires reasoning not only about what is said, but also about what is left unsaid. We propose RADAR, a role-anchored multi-agent debate framework for omission-aware fact verification under realistic, noisy retrieval. RADAR assigns complementary roles to a Politician and a Scientist, who reason adversarially over shared retrieved evidence, moderated by a neutral Judge. A dual-threshold early termination controller adaptively decides when sufficient reasoning has been reached to issue a verdict. Experiments show that RADAR consistently outperforms strong single- and multi-agent baselines across datasets and backbones, improving omission detection accuracy while reducing reasoning cost. These results demonstrate that role-anchored, retrieval-grounded debate with adaptive control is an effective and scalable framework for uncovering missing context in fact verification. The code is available at https://github.com/tangyixuan/RADAR.

cs.CL↗

arXiv2026

The Power of Information for Intermediate States in Contract Design

In the conventional principal-agent problem, a principal delegates a task to an agent and formulates a contract to incentivize the agent's actions on behalf of the principal. However, this framework overlooks the information that is possibly available during the delegation process in some scenarios. To address this limitation, we propose a novel model that incorporates multiple intermediate states to capture such information revealed during the delegation. Furthermore, to evaluate the impact of the information embedded in these intermediate states, we introduce two distinct contracts: the pay-halfway contract, which provides payments based not only on final outcomes but also on intermediate states, and the terminate-halfway contract, which allows the principal to terminate the delegation process upon encountering undesirable intermediate states. This leads to the question of whether and how these contract types can leverage intermediate-state information? In particular, we ask: Can these contract types outperform standard contracts, and if so, when and to what extent? We answer the first question affirmatively and provide several important insights regarding the second, shedding light on the circumstances in which intermediate-state-aware contracts yield substantial advantages.

cs.GT↗

arXiv2025

PROMISE: Prompt-Attentive Hierarchical Contrastive Learning for Robust Cross-Modal Representation with Missing Modalities

Multimodal models integrating natural language and visual information have substantially improved generalization of representation models. However, their effectiveness significantly declines in real-world situations where certain modalities are missing or unavailable. This degradation primarily stems from inconsistent representation learning between complete multimodal data and incomplete modality scenarios. Existing approaches typically address missing modalities through relatively simplistic generation methods, yet these approaches fail to adequately preserve cross-modal consistency, leading to suboptimal performance. To overcome this limitation, we propose a novel multimodal framework named PROMISE, a PROMpting-Attentive HIerarchical ContraStive LEarning approach designed explicitly for robust cross-modal representation under conditions of missing modalities. Specifically, PROMISE innovatively incorporates multimodal prompt learning into a hierarchical contrastive learning framework, equipped with a specially designed prompt-attention mechanism. This mechanism dynamically generates robust and consistent representations for scenarios where particular modalities are absent, thereby effectively bridging the representational gap between complete and incomplete data. Extensive experiments conducted on benchmark datasets, along with comprehensive ablation studies, clearly demonstrate the superior performance of PROMISE compared to current state-of-the-art multimodal methods.

cs.CV↗

arXiv2025

BaThe: Defense against the Jailbreak Attack in Multimodal Large Language Models by Treating Harmful Instruction as Backdoor Trigger

Multimodal Large Language Models (MLLMs) have showcased impressive performance in a variety of multimodal tasks. On the other hand, the integration of additional image modality may allow the malicious users to inject harmful content inside the images for jailbreaking. Unlike text-based LLMs, where adversaries need to select discrete tokens to conceal their malicious intent using specific algorithms, the continuous nature of image signals provides a direct opportunity for adversaries to inject harmful intentions. In this work, we propose $\textbf{BaThe}$ ($\textbf{Ba}$ckdoor $\textbf{T}$rigger S$\textbf{h}$i$\textbf{e}$ld), a simple yet effective jailbreak defense mechanism. Our work is motivated by recent research on jailbreak backdoor attack and virtual prompt backdoor attack in generative language models. Jailbreak backdoor attack uses harmful instructions combined with manually crafted strings as triggers to make the backdoored model generate prohibited responses. We assume that harmful instructions can function as triggers, and if we alternatively set rejection responses as the triggered response, the backdoored model then can defend against jailbreak attacks. We achieve this by utilizing virtual rejection prompt, similar to the virtual prompt backdoor attack. We embed the virtual rejection prompt into the soft text embeddings, which we call ``wedge''. Our comprehensive experiments demonstrate that BaThe effectively mitigates various types of jailbreak attacks and is adaptable to defend against unseen attacks, with minimal impact on MLLMs' performance.

cs.CR↗

arXiv2025

Continuous non-equilibrium transition driven by the heat flow

We discovered an out-of-equilibrium transition in the ideal gas between two walls, divided by an inner, adiabatic, movable wall. The system is driven out-of-equilibrium by supplying energy directly into the volume of the gas. At critical heat flux, we have found a continuous transition to the state with a low-density, hot gas on one side of the movable wall and a dense, cold gas on the other side. Molecular dynamic simulations of the soft-sphere fluid confirm the existence of the transition in the interacting system. We introduce a stationary state Helmholtz-like function whose minimum determines the stable positions of the internal wall. This transition can be used as a paradigm of transitions in stationary states and the Helmholtz-like function as a paradigm of the thermodynamic description of these states.

cond-mat.stat-mech↗

arXiv2024

Atmospheric Pressure Ammonia Synthesis on AuRu Catalysts Enabled by Plasmon-Controlled Hydrogenation and Nitrogen-species Desorption

Ammonia is a key component of fertilizer and a potential clean fuel and hydrogen carrier. The Haber-Bosch process for ammonia synthesis consumes more than half of industrial hydrogen and contributes up to ~3% of global greenhouse gas emissions. Light-driven reactions via surface plasmon resonances offer a less energy-intensive pathway for ammonia production by altering reaction intermediates. Here, we report gold-ruthenium plasmonic bimetallic alloys for ammonia synthesis at room temperature and pressure, driven by visible light. We use colloidal synthesis to create AuRu$_x$ alloys (x=0.1, 0.2, 0.3) and disperse these nanoparticles on MgO supports for gas-phase ammonia synthesis. We observe a ~60 $μ$mol/g/h reactivity and ~0.12% external quantum efficiency on a AuRu$_0$$_.$$_2$ sample under 100 mW/cm$^2$ visible light. In-situ diffuse reflective infrared Fourier transform spectroscopic measurements show that hydrogenation of nitrogen adsorbates is accelerated under light compared to thermocatalysis. Combining wavelength-dependent reactivity and spectroscopic findings with semi-classical electromagnetic modeling, we show plasmonic bimetallic alloys expedite ammonia synthesis by aiding hydrogenation of adsorbed nitrogen species via plasmon-mediated hot electrons. Quantum mechanical calculations reveal hydrogen-assisted N$_2$ splitting in the excited state is key to activating the reaction under ambient conditions. Therefore, light or H$_2$ alone cannot dissociate N$_2$ -- the key bottleneck to breaking N$_2$'s triple bond. Our findings are consistent with recent hypotheses on how nitrogenase enzymes catalyze ammonia production at mild conditions and provide insights for sustainable photochemical transformations.

physics.chem-ph↗

arXiv2023

Inferring geometrical dynamics of cell nucleus translocation

The ability of eukaryotic cells to squeeze through constrictions is limited by the stiffness of their large and rigid nucleus. However, migrating cells are often able to overcome this limitation and pass through constrictions much smaller than their nucleus, a mechanism that is not yet understood. This is what we address here through a data-driven approach using microfluidic devices where cells migrate through controlled narrow spaces of sizes comparable to the ones encountered in physiological situations. Stochastic Force Inference is applied to experimental nuclear trajectories and nuclear shape descriptors, resulting in equations that effectively describe this phenomenon of nuclear translocation. By employing a model where the channel geometry is an explicit parameter and by training it over experimental data with different sizes of constrictions, we ensure that the resulting equations are predictive to other geometries. Altogether, the approach developed here paves the way for a mechanistic and quantitative description of dynamical cell complexity during its motility.

physics.bio-ph↗

arXiv2023

ChatLLM Network: More brains, More intelligence

Dialogue-based language models mark a huge milestone in the field of artificial intelligence, by their impressive ability to interact with users, as well as a series of challenging tasks prompted by customized instructions. However, the prevalent large-scale dialogue-based language models like ChatGPT still have room for improvement, such as unstable responses to questions and the inability to think cooperatively like humans. Considering the ability of dialogue-based language models in conversation and their inherent randomness in thinking, we propose ChatLLM network that allows multiple dialogue-based language models to interact, provide feedback, and think together. We design the network of ChatLLMs based on ChatGPT. Specifically, individual instances of ChatGPT may possess distinct perspectives towards the same problem, and by consolidating these diverse viewpoints via a separate ChatGPT, the ChatLLM network system can conduct decision-making more objectively and comprehensively. In addition, a language-based feedback mechanism comparable to backpropagation is devised to update the ChatGPTs within the network. Experiments on two datasets demonstrate that our network attains significant improvements in problem-solving, leading to observable progress amongst each member.

cs.AI↗

arXiv2021

Internal energy in compressible Poiseuille flow

We analyse a compressible Poiseuille flow of ideal gas in a plane channel. We provide the form of internal energy U for a non-equilibrium stationary state (NESS) that includes viscous dissipation and pressure work. We demonstrate that U depends strongly on the ratio Δp/p_0, where Δp is the pressure difference between inlet and outlet and p_0 is the outlet's pressure. In addition, U depends on two other variables: the channel aspect ratio and the parameter equivalent to Reynolds number. The stored internal energy, ΔU=U-U0, is small compared to the internal energy U0 of the equilibrium state (ES) for a moderate range of values of Δp/p_0. However, ΔU can become large for big Δp or close to vacuum conditions at the outlet (p_0~0 Pa).

physics.flu-dyn↗

arXiv2020

Theory of coupled ion-electron transfer kinetics

The microscopic theory of chemical reactions is based on transition state theory, where atoms or ions transfer classically over an energy barrier, as electrons maintain their ground state. Electron transfer is fundamentally different and occurs by tunneling in response to solvent fluctuations. Here, we develop the theory of coupled ion-electron transfer, in which ions and solvent molecules fluctuate cooperatively to facilitate electron transfer. We derive a general formula of the reaction rate that depends on the overpotential, solvent properties, the electronic structure of the electron donor/acceptor, and the excess chemical potential of ions in the transition state. For Faradaic reactions, the theory predicts curved Tafel plots with a concentration-dependent reaction-limited current. For moderate overpotentials, our formula reduces to the Butler-Volmer equation and explains its relevance, not only in the well-known limit of large electron-transfer (solvent reorganization) energy, but also in the opposite limit of large ion-transfer energy. The rate formula is applied to Li-ion batteries, where reduction of the electrode host material couples with ion insertion. In the case of lithium iron phosphate, the theory accurately predicts the concentration dependence of the exchange current measured by {\it in operando} X-Ray microscopy without any adjustable parameters. These results pave the way for interfacial engineering to enhance ion intercalation rates, not only for batteries, but also for ionic separations and neuromorphic computing.

physics.chem-ph↗

arXiv2020

Storage of energy in constrained non-equilibrium systems

We study a quantity $\mathcal{T}$ defined as the energy U, stored in non-equilibrium steady states (NESS) over its value in equilibrium $U_0$, $ΔU=U-U_0$ divided by the heat flow $J_{U}$ going out of the system. A recent study suggests that $\mathcal{T}$ is minimized in steady states (Phys.Rev.E.99, 042118 (2019)). We evaluate this hypothesis using an ideal gas system with three methods of energy delivery: from a uniformly distributed energy source, from an external heat flow through the surface, and from an external matter flow. By introducing internal constraints into the system, we determine $\mathcal{T}$ with and without constraints and find that $\mathcal{T}$ is the smallest for unconstrained NESS. We find that the form of the internal energy in the studied NESS follows $U=U_0*f(J_U)$. In this context, we discuss natural variables for NESS, define the embedded energy (an analog of Helmholtz free energy for NESS), and provide its interpretation.

cond-mat.stat-mech↗

arXiv2020

Energy Storage in Steady States under Cyclic Local Energy Input

We study periodic steady states of a lattice system under external cyclic energy supply using simulation. We consider different protocols for cyclic energy supply and examine the energy storage. Under the same energy flux, we found that the stored energy depends on the details of the supply, period and amplitude of the supply. Further, we introduce an adiabatic wall as internal constrain into the lattice and examine the stored energy with respect to different positions of the internal constrain. We found that the stored energy for constrained systems are larger than their unconstrained counterpart. We also observe that the system stores more energy through large and rare energy delivery, comparing to small and frequent delivery.

cond-mat.stat-mech↗

arXiv2019

Flux and storage of energy in non-equilibrium, stationary states

Systems kept out of equilibrium in stationary states by an external source of energy store an energy $ΔU=U-U_0$. $U_0$ is the internal energy at equilibrium state, obtained after the shutdown of energy input. We determine $ΔU$ for two model systems: ideal gas and Lennard-Jones fluid. $ΔU$ depends not only on the total energy flux, $J_U$, but also on the mode of energy transfer into the system. We use three different modes of energy transfer where: the energy flux per unit volume is (i) constant; (ii) proportional to the local temperature (iii) proportional to the local density. We show that $ΔU /J_U=τ$ is minimized in the stationary states formed in these systems, irrespective of the mode of energy transfer. $τ$ is the characteristic time scale of energy outflow from the system immediately after the shutdown of energy flux. We prove that $τ$ is minimized in stable states of the Rayleigh-Benard cell.

cond-mat.stat-mech↗

arXiv2016

Critical behavior of entropy production and learning rate: Ising model with an oscillating field

We study the critical behavior of the entropy production of the Ising model subject to a magnetic field that oscillates in time. The mean-field model displays a phase transition that can be either first or second-order, depending on the amplitude of the field and on the frequency of oscillation. Within this approximation the entropy production rate is shown to have a discontinuity when the transition is first-order and to be continuous, with a jump in its first derivative, if the transition is second-order. In two dimensions, we find with numerical simulations that the critical behavior of the entropy production rate is the same, independent of the frequency and amplitude of the field. Its first derivative has a logarithmic divergence at the critical point. This result is in agreement with the lack of a first-order phase transition in two dimensions. We analyze a model with a field that changes at stochastic time-intervals between two values. This model allows for an informational theoretic interpretation, with the system as a sensor that follows the external field. We calculate numerically a lower bound on the learning rate, which quantifies how much information the system obtains about the field. Its first derivative with respect to temperature is found to have a jump at the critical point.

cond-mat.stat-mech↗