arXiv subjects

Jiaxin Zhang

Publications and source records attributed to Jiaxin Zhang.

At least 55 records · Page 3Linked to original sources

arXiv2026

Nudging the Boundaries of LLM Reasoning

Current online reinforcement learning (RL) algorithms like GRPO share a key limitation in LLM reasoning: they cannot learn from problems that are "unsolvable" to the model. In other words, they can only improve performance on problems where the model is capable of exploring the correct answer. Consequently, the model's "upper limit" remains unchanged after RL training, even though the likelihood of solving easier, solvable problems may increase. These hard samples cannot contribute to training, as no rollouts yield rewards and thus no gradients are produced. To unlock learning from these hard samples, we propose NuRL, a "nudging" method that aims to push the upper bound of LLM reasoning using self-generated hints, i.e., abstract cues that help reduce the problem difficulty for the model. Given a question and its gold answer, the model generates a CoT and then produces a hint containing the core knowledge needed to solve the problem. During training, we generate G rollouts from the base policy and use the pass rate to decide whether the hint should be injected. For hard samples with a 0% pass rate, we inject the hint and regenerate a new batch of trajectories. This yields two benefits: (1) the hint boosts pass rates (from 0% to non-zero), thereby introducing training signals for previously unsolvable samples, and (2) the hints are self-generated, avoiding distributional shift and do not rely on external models. NuRL achieves consistent improvements across 6 benchmarks and 3 models, while remaining complementary to test-time scaling. Notably, NuRL can raise the model's upper limit, whereas GRPO leaves pass@1024 unchanged from the base model. Furthermore, we present a systematic study of what makes an effective hint and when hints are most useful. Interestingly, the best hints are abstract and high-level, and are most beneficial when applied necessarily and after GRPO has converged.

cs.LG↗

arXiv2026

ARGOS: Automated Functional Safety Requirement Synthesis for Embodied AI via Attribute-Guided Combinatorial Reasoning

Ensuring functional safety is essential for the deployment of Embodied AI in complex open-world environments. However, traditional Hazard Analysis and Risk Assessment (HARA) methods struggle to scale in this domain. While HARA relies on enumerating risks for finite and pre-defined function lists, Embodied AI operates on open-ended natural language instructions, creating a challenge of combinatorial interaction risks. Whereas Large Language Models (LLMs) have emerged as a promising solution to this scalability challenge, they often lack physical grounding, yielding semantically superficial and incoherent hazard descriptions. To overcome these limitations, we propose a new framework ARGOS (AttRibute-Guided cOmbinatorial reaSoning), which bridges the gap between open-ended user instructions and concrete physical attributes. By dynamically decomposing entities from instructions into these fine-grained properties, ARGOS grounds LLM reasoning in causal risk factors to generate physically plausible hazard scenarios. It then instantiates abstract safety standards, such as ISO 13482, into context-specific Functional Safety Requirements (FSRs) by integrating these scenarios with robot capabilities. Extensive experiments validate that ARGOS produces high-quality FSRs and outperforms baselines in identifying long-tail risks. Overall, this work paves the way for systematic and grounded functional safety requirement generation, a critical step toward the safe industrial deployment of Embodied AI.

cs.RO↗

arXiv2026

Beyond Parameter Finetuning: Test-Time Representation Refinement for Node Classification

Graph Neural Networks frequently exhibit significant performance degradation in the out-of-distribution test scenario. While test-time training (TTT) offers a promising solution, existing Parameter Finetuning (PaFT) paradigm suffer from catastrophic forgetting, hindering their real-world applicability. We propose TTReFT, a novel Test-Time Representation FineTuning framework that transitions the adaptation target from model parameters to latent representations. Specifically, TTReFT achieves this through three key innovations: (1) uncertainty-guided node selection for specific interventions, (2) low-rank representation interventions that preserve pre-trained knowledge, and (3) an intervention-aware masked autoencoder that dynamically adjust masking strategy to accommodate the node selection scheme. Theoretically, we establish guarantees for TTReFT in OOD settings. Empirically, extensive experiments across five benchmark datasets demonstrate that TTReFT achieves consistent and superior performance. Our work establishes representation finetuning as a new paradigm for graph TTT, offering both theoretical grounding and immediate practical utility for real-world deployment.

cs.LG↗

arXiv2026

Agentic Uncertainty Quantification

Although AI agents have demonstrated impressive capabilities in long-horizon reasoning, their reliability is severely hampered by the ``Spiral of Hallucination,'' where early epistemic errors propagate irreversibly. Existing methods face a dilemma: uncertainty quantification (UQ) methods typically act as passive sensors, only diagnosing risks without addressing them, while self-reflection mechanisms suffer from continuous or aimless corrections. To bridge this gap, we propose a unified Dual-Process Agentic UQ (AUQ) framework that transforms verbalized uncertainty into active, bi-directional control signals. Our architecture comprises two complementary mechanisms: System 1 (Uncertainty-Aware Memory, UAM), which implicitly propagates verbalized confidence and semantic explanations to prevent blind decision-making; and System 2 (Uncertainty-Aware Reflection, UAR), which utilizes these explanations as rational cues to trigger targeted inference-time resolution only when necessary. This enables the agent to balance efficient execution and deep deliberation dynamically. Extensive experiments on closed-loop benchmarks and open-ended deep research tasks demonstrate that our training-free approach achieves superior performance and trajectory-level calibration. We believe this principled framework AUQ represents a significant step towards reliable agents.

cs.AI↗

arXiv2026

Agentic Confidence Calibration

AI agents are rapidly advancing from passive language models to autonomous systems executing complex, multi-step tasks. Yet their overconfidence in failure remains a fundamental barrier to deployment in high-stakes settings. Existing calibration methods, built for static single-turn outputs, cannot address the unique challenges of agentic systems, such as compounding errors along trajectories, uncertainty from external tools, and opaque failure modes. To address these challenges, we introduce, for the first time, the problem of Agentic Confidence Calibration and propose Holistic Trajectory Calibration (HTC), a novel diagnostic framework that extracts rich process-level features ranging from macro dynamics to micro stability across an agent's entire trajectory. Powered by a simple, interpretable model, HTC consistently surpasses strong baselines in both calibration and discrimination, across eight benchmarks, multiple LLMs, and diverse agent frameworks. Beyond performance, HTC delivers three essential advances: it provides interpretability by revealing the signals behind failure, enables transferability by applying across domains without retraining, and achieves generalization through a General Agent Calibrator (GAC) that achieves the best calibration (lowest ECE) on the out-of-domain GAIA benchmark. Together, these contributions establish a new process-centric paradigm for confidence calibration, providing a framework for diagnosing and enhancing the reliability of AI agents.

cs.AI↗

arXiv2025

Multi-Modal Feature Fusion for Spatial Morphology Analysis of Traditional Villages via Hierarchical Graph Neural Networks

Villages areas hold significant importance in the study of human-land relationships. However, with the advancement of urbanization, the gradual disappearance of spatial characteristics and the homogenization of landscapes have emerged as prominent issues. Existing studies primarily adopt a single-disciplinary perspective to analyze villages spatial morphology and its influencing factors, relying heavily on qualitative analysis methods. These efforts are often constrained by the lack of digital infrastructure and insufficient data. To address the current research limitations, this paper proposes a Hierarchical Graph Neural Network (HGNN) model that integrates multi-source data to conduct an in-depth analysis of villages spatial morphology. The framework includes two types of nodes-input nodes and communication nodes-and two types of edges-static input edges and dynamic communication edges. By combining Graph Convolutional Networks (GCN) and Graph Attention Networks (GAT), the proposed model efficiently integrates multimodal features under a two-stage feature update mechanism. Additionally, based on existing principles for classifying villages spatial morphology, the paper introduces a relational pooling mechanism and implements a joint training strategy across 17 subtypes. Experimental results demonstrate that this method achieves significant performance improvements over existing approaches in multimodal fusion and classification tasks. Additionally, the proposed joint optimization of all sub-types lifts mean accuracy/F1 from 0.71/0.83 (independent models) to 0.82/0.90, driven by a 6% gain for parcel tasks. Our method provides scientific evidence for exploring villages spatial patterns and generative logic.

cs.CV↗

arXiv2025

Multiple radial SLE(0) and classical Calogero-Sutherland System

We develop a theory of multiple radial SLE(0) -- a smooth system of curves in a simply connected domain $Ω$ with marked boundary points $z_1, \ldots, z_n \in \partial Ω$ and a marked interior point $q$ -- arising as the deterministic limit of random multiple radial SLE($κ$) systems. We construct multiple radial SLE(0) systems by starting from the stationary relations, which arise heuristically as the $κ\to 0$ limit of partition functions. By constructing the field integrals of motion for the Loewner dynamics, we show that the traces of multiple radial SLE(0) systems are the horizontal trajectories of an equivalence class of quadratic differentials. These trajectories have limiting ends at the boundary points $\{z_1, z_2, \ldots, z_n\}$. The stationary relations connect the classification of multiple radial SLE(0) systems to the enumeration of critical points of the master function of trigonometric Knizhnik--Zamolodchikov (KZ) equations. In the deterministic case $κ= 0$, we show that the Loewner dynamics with a common parametrization of capacity form a special class of classical Calogero--Sutherland systems, restricted to a submanifold of phase space defined by the Lax matrix.

math.PR↗

arXiv2025

Multiple radial SLE($κ$) and quantum Calogero-Sutherland system

We develop a theory for the multiple radial $\mathrm{SLE}(κ)$ systems with parameter $κ> 0$ -- a family of random multi-curve systems in a simply connected domain $Ω$, with marked boundary points $z_1, \ldots, z_n \in \partial Ω$ and a marked interior point $q$. As a consequence of the domain Markov property and conformal invariance, we show that such systems are characterized by equivalence classes of partition functions, which are not necessarily conformally covariant. Nevertheless, within each equivalence class, one can always choose a conformally covariant representative. When $Ω$ is taken to be the unit disk $\mathbb{D}$ and the marked interior point $q$ is set at the origin, we demonstrate that the partition function satisfies a system of second-order PDEs, known as the null vector equations, with a null vector constant $h$ and a rotation equation involving a constant $ω$. Motivated by the Coulomb gas formalism in conformal field theory, we construct four families of solutions to the null vector equations, which are naturally classified according to topological link patterns. For $κ> 0$, the partition functions of multiple radial $\mathrm{SLE}(κ)$ systems correspond to eigenstates of the quantum Calogero--Sutherland (CS) Hamiltonian beyond the states built upon the fermionic states.

math.PR↗

arXiv2025

GAUSS: Benchmarking Structured Mathematical Skills for Large Language Models

We introduce \textbf{GAUSS} (\textbf{G}eneral \textbf{A}ssessment of \textbf{U}nderlying \textbf{S}tructured \textbf{S}kills in Mathematics), a benchmark that evaluates LLMs' mathematical abilities across twelve core skill dimensions, grouped into three domains: knowledge and understanding, problem solving and communication, and meta-skills and creativity. By categorizing problems according to cognitive skills and designing tasks that isolate specific abilities, GAUSS constructs comprehensive, fine-grained, and interpretable profiles of models' mathematical abilities. These profiles faithfully represent their underlying mathematical intelligence. To exemplify how to use the \textsc{GAUSS} benchmark, we have derived the skill profile of \textsc{GPT-5-thinking}, revealing its strengths and weaknesses as well as its differences relative to \textsc{o4-mini-high}, thereby underscoring the value of multidimensional, skill-based evaluation.

cs.AI↗

arXiv2025

CookBench: A Long-Horizon Embodied Planning Benchmark for Complex Cooking Scenarios

Embodied Planning is dedicated to the goal of creating agents capable of executing long-horizon tasks in complex physical worlds. However, existing embodied planning benchmarks frequently feature short-horizon tasks and coarse-grained action primitives. To address this challenge, we introduce CookBench, a benchmark for long-horizon planning in complex cooking scenarios. By leveraging a high-fidelity simulation environment built upon the powerful Unity game engine, we define frontier AI challenges in a complex, realistic environment. The core task in CookBench is designed as a two-stage process. First, in Intention Recognition, an agent needs to accurately parse a user's complex intent. Second, in Embodied Interaction, the agent should execute the identified cooking goal through a long-horizon, fine-grained sequence of physical actions. Unlike existing embodied planning benchmarks, we refine the action granularity to a spatial level that considers crucial operational information while abstracting away low-level robotic control. Besides, We provide a comprehensive toolset that encapsulates the simulator. Its unified API supports both macro-level operations, such as placing orders and purchasing ingredients, and a rich set of fine-grained embodied actions for physical interaction, enabling researchers to focus on high-level planning and decision-making. Furthermore, we present an in-depth analysis of state-of-the-art, closed-source Large Language Model and Vision-Language Model, revealing their major shortcomings and challenges posed by complex, long-horizon tasks. The full benchmark will be open-sourced to facilitate future research.

cs.RO↗

arXiv2025

SEE: Strategic Exploration and Exploitation for Cohesive In-Context Prompt Optimization

Designing optimal prompts for Large Language Models (LLMs) is a complicated and resource-intensive task, often requiring substantial human expertise and effort. Existing approaches typically separate the optimization of prompt instructions and in-context learning examples, leading to incohesive prompts that are defined and represented by suboptimal task performance. To overcome these challenges, we propose a novel Cohesive In-Context Prompt Optimization framework that refines both prompt instructions and examples. However, formulating such an optimization in the discrete and high-dimensional space of natural language poses significant challenges in both convergence and computational efficiency. To address these issues, we introduce SEE, a scalable and efficient prompt optimization framework that adopts metaheuristic optimization principles and strategically balances exploration and exploitation to enhance optimization performance and achieve efficient convergence. SEE features a quad-phased design that alternates between global traversal (exploration) and local optimization (exploitation) and adaptively chooses LLM operators during the optimization process. We have conducted a comprehensive evaluation across 35 benchmark tasks, and SEE significantly outperforms state-of-the-art baseline methods by a large margin, achieving an average performance gain of 13.94 while reducing computational costs by 58.67.

cs.CL↗

arXiv2025

A Survey of Automatic Prompt Optimization with Instruction-focused Heuristic-based Search Algorithm

Recent advances in Large Language Models have led to remarkable achievements across a variety of Natural Language Processing tasks, making prompt engineering increasingly central to guiding model outputs. While manual methods can be effective, they typically rely on intuition and do not automatically refine prompts over time. In contrast, automatic prompt optimization employing heuristic-based search algorithms can systematically explore and improve prompts with minimal human oversight. This survey proposes a comprehensive taxonomy of these methods, categorizing them by where optimization occurs, what is optimized, what criteria drive the optimization, which operators generate new prompts, and which iterative search algorithms are applied. We further highlight specialized datasets and tools that support and accelerate automated prompt refinement. We conclude by discussing key open challenges pointing toward future opportunities for more robust and versatile LLM applications.

cs.CL↗

arXiv2025

Gradient-guided Attention Map Editing: Towards Efficient Contextual Hallucination Mitigation

In tasks like summarization and open-book question answering (QA), Large Language Models (LLMs) often encounter "contextual hallucination", where they produce irrelevant or incorrect responses despite having access to accurate source information. This typically occurs because these models tend to prioritize self-generated content over the input context, causing them to disregard pertinent details. To address this challenge, we introduce a novel method called "Guided Attention Map Editing" (GAME), which dynamically adjusts attention maps to improve contextual relevance. During inference, GAME employs a trained classifier to identify attention maps prone to inducing hallucinations and executes targeted interventions. These interventions, guided by gradient-informed "edit directions'', strategically redistribute attention weights across various heads to effectively reduce hallucination. Comprehensive evaluations on challenging summarization and open-book QA tasks show that GAME consistently reduces hallucinations across a variety of open-source models. Specifically, GAME reduces hallucinations by 10% in the XSum summarization task while achieving a 7X speed-up in computational efficiency compared to the state-of-the-art baselines.

cs.CL↗

arXiv2025

Reconfigurable non-Abelian geometric phase in hybrid integrated photonics

The non-Abelian geometric phase possesses the capability of enabling robust and fault-resilient unitary transformations, making it a cornerstone of holonomic quantum computation. This "all-geometric" approach has successfully advanced the manipulation of electrons in condensed matter physics and has sparked growing interest in its implementation within photonics, an area that has traditionally relied on sensitive dynamic phases. However, a major limitation of the topologically protected and inherently robust geometric phase is its lack of reconfigurability. In contrast, mainstream optical computing schemes demand high reconfigurability to compensate for fabrication errors and to support diverse computational tasks. Here, we demonstrate a reconfigurable non-Abelian geometric phase based on the non-volatile phase-change material Sb$_2$Se$_3$. By switching between its crystalline and amorphous states, the number of degenerate subspaces can be actively adjusted. Thus, multilevel second-order matrices and reconfigurable third-order matrices with 3-bit control is realized. For larger reconfigurable rotation angles, tunable braiding operations are also demonstrated. Furthermore, high-dimensional reconfigurable braiding shows promising potential for applications in optical switching. Our results pave the way for the all-geometric-phase-based approach in optical computing.

physics.optics↗

arXiv2025

From System 1 to System 2: A Survey of Reasoning Large Language Models

Achieving human-level intelligence requires refining the transition from the fast, intuitive System 1 to the slower, more deliberate System 2 reasoning. While System 1 excels in quick, heuristic decisions, System 2 relies on logical reasoning for more accurate judgments and reduced biases. Foundational Large Language Models (LLMs) excel at fast decision-making but lack the depth for complex reasoning, as they have not yet fully embraced the step-by-step analysis characteristic of true System 2 thinking. Recently, reasoning LLMs like OpenAI's o1/o3 and DeepSeek's R1 have demonstrated expert-level performance in fields such as mathematics and coding, closely mimicking the deliberate reasoning of System 2 and showcasing human-like cognitive abilities. This survey begins with a brief overview of the progress in foundational LLMs and the early development of System 2 technologies, exploring how their combination has paved the way for reasoning LLMs. Next, we discuss how to construct reasoning LLMs, analyzing their features, the core methods enabling advanced reasoning, and the evolution of various reasoning LLMs. Additionally, we provide an overview of reasoning benchmarks, offering an in-depth comparison of the performance of representative reasoning LLMs. Finally, we explore promising directions for advancing reasoning LLMs and maintain a real-time \href{https://github.com/zzli2022/Awesome-Slow-Reason-System}{GitHub Repository} to track the latest developments. We hope this survey will serve as a valuable resource to inspire innovation and drive progress in this rapidly evolving field.

cs.AI↗

arXiv2025

Irregular traces of multiple SLE(0) systems with multiple marked points

In this supplementary note, we study the traces of multiple SLE(0) systems with two or more additional marked points. For general chordal configurations, the traces correspond to the real locus of real rational functions; in the radial case, they correspond to the horizontal trajectories of residue-free quadratic differentials. In both settings, we establish the regularity of the trajectories near singularities: no spiraling occurs, and no two trajectories asymptotically converge to the same direction. Moreover, in the radial case with non-zero spin at the marked interior point, we show that the spin induces a spiraling behavior at the marked interior point. However, this regularity breaks down when multiple interior marked points are present. In such cases, trajectories may asymptotically approach the same direction, and spiraling can occur even in the absence of spin. We present explicit counterexamples generated using MATLAB, with code provided for reference.

math.PR↗

arXiv2025

Modality-Specialized Synergizers for Interleaved Vision-Language Generalists

Recent advancements in Vision-Language Models (VLMs) have led to the emergence of Vision-Language Generalists (VLGs) capable of understanding and generating both text and images. However, seamlessly generating an arbitrary sequence of text and images remains a challenging task for the current VLGs. One primary limitation lies in applying a unified architecture and the same set of parameters to simultaneously model discrete text tokens and continuous image features. Recent works attempt to tackle this fundamental problem by introducing modality-aware expert models. However, they employ identical architectures to process both text and images, disregarding the intrinsic inductive biases in these two modalities. In this work, we introduce MODALITY-SPECIALIZED SYNERGIZERS (MOSS), a novel design that efficiently optimizes existing unified architectures of VLGs with modality-specialized adaptation layers, i.e., a Convolutional LoRA for modeling the local priors of image patches and a Linear LoRA for processing sequential text. This design enables more effective modeling of modality-specific features while maintaining the strong cross-modal integration gained from pretraining. In addition, to improve the instruction-following capability on interleaved text-and-image generation, we introduce LEAFINSTRUCT, the first open-sourced interleaved instruction tuning dataset comprising 184,982 high-quality instances on more than 10 diverse domains. Extensive experiments show that VLGs integrated with M OSS achieve state-of-the-art performance, significantly surpassing baseline VLGs in complex interleaved generation tasks. Furthermore, our method exhibits strong generalizability on different VLGs.

cs.CL↗

arXiv2025

Multiple chordal SLE($κ$) and quantum Calogero-Moser system

We study multiple chordal SLE$(κ)$ systems in a simply connected domain $Ω$, where $z_1, \ldots, z_n \in \partial Ω$ are boundary starting points and $q \in \partial Ω$ is an additional marked boundary point. As a consequence of the domain Markov property and conformal invariance, we show that the presence of the marked boundary point $q$ gives rise to a natural equivalence relation on partition functions. While these functions are not necessarily conformally covariant, each equivalence class contains a conformally covariant representative. Building on the framework introduced in \cite{Dub07}, we demonstrate that in the $\mathbb{H}$-uniformization with $q = \infty$, the partition functions satisfy both the null vector equations and a dilatation equation with scaling exponent $d$. Using techniques from the Coulomb gas formalism in conformal field theory, we construct two distinct families of solutions, each indexed by a topological link pattern of type $(n, m)$ with $2m \leq n$. In the special case $Ω= \mathbb{H}$ and $q = \infty$, we further show that these partition functions correspond to eigenstates of the quantum Calogero-Moser system, thereby extending the known correspondence beyond the standard $(2n, n)$ setting.

math.PR↗