Searcharxiv⌕ Search

arXiv subjects

Haoyang Chen

Publications and source records attributed to Haoyang Chen.

At least 37 records · Page 2Linked to original sources
arXiv2026

Editable XAI: Toward Bidirectional Human-AI Alignment with Co-Editable Explanations of Interpretable Attributes

While Explainable AI (XAI) helps users understand AI decisions, misalignment in domain knowledge can lead to disagreement. This inconsistency hinders understanding, and because explanations are often read-only, users lack the control to improve alignment. We propose making XAI editable, allowing users to write rules to improve control and gain deeper understanding through the generation effect of active learning. We developed CoExplain, leveraging a neural network for universal representation and symbolic rules for intuitive reasoning on interpretable attributes. CoExplain explains the neural network with a faithful proxy decision tree, parses user-written rules as an equivalent neural network graph, and collaboratively optimizes the decision tree. In a user study (N=43), CoExplain and manually editable XAI improved user understanding and model alignment compared to read-only XAI. CoExplain was easier to use with fewer edits and less time. This work contributes Editable XAI for bidirectional AI alignment, improving understanding and control.

cs.HC↗
arXiv2026

Low regularity ill-posedness and shock formation for 3D ideal compressible MHD

The study of magnetohydrodynamics (MHD) significantly boosts the understanding and development of solar physics, planetary dynamics and controlled nuclear fusion. Dynamical properties of the MHD system involve nonlinear interactions of waves with multiple travelling speeds (the fast and slow magnetosonic waves, the Alfvén wave and the entropy wave). One intriguing topic is the shock phenomena accompanied by the magnetic field, which have been affirmed by astronomical observations. However, permitting the residence of all above multi-speed waves, mathematically, whether one can prove shock formation for three dimensional (3D) MHD is still open. The multiple-speed nature of the MHD system makes it fascinating and challenging. In this paper, we report our recent progress in answering the above question. For 3D ideal compressible MHD, we construct planar symmetric examples of shock formation allowing the presence of all characteristic waves with multiple wave speeds. This also answers a question raised by Majda on conservation law in 1984. Building on our construction, we further prove that the Cauchy problem for 3D ideal MHD is $H^2$ ill-posed. And this is caused by the shock formation. In particular, when the magnetic field is absent, we also provide a desired low-regularity ill-posedness result for the 3D compressible Euler equations, and it is sharp with respect to the regularity of the fluid velocity. Our proof for 3D MHD is based on a coalition of a carefully designed algebraic approach and a geometric approach. To trace the nonlinear interactions of various waves, we algebraically decompose the 3D ideal MHD equations into a $7\times 7$ non-strictly hyperbolic system. Via detailed calculations, we reveal its hidden subtle structures. With them we give a complete description of MHD dynamics up to the earliest singular event, when a shock forms.

math.AP↗
arXiv2026

A neural network for modeling human concept formation, understanding and communication

A remarkable capability of the human brain is to form more abstract conceptual representations from sensorimotor experiences and flexibly apply them independent of direct sensory inputs. However, the computational mechanism underlying this ability remains poorly understood. Here, we present a dual-module neural network framework, the CATS Net, to bridge this gap. Our model consists of a concept-abstraction module that extracts low-dimensional conceptual representations, and a task-solving module that performs visual judgement tasks under the hierarchical gating control of the formed concepts. The system develops transferable semantic structure based on concept representations that enable cross-network knowledge transfer through conceptual communication. Model-brain fitting analyses reveal that these emergent concept spaces align with both neurocognitive semantic model and brain response structures in the human ventral occipitotemporal cortex, while the gating mechanisms mirror that in the semantic control brain network. This work establishes a unified computational framework that can offer mechanistic insights for understanding human conceptual cognition and engineering artificial systems with human-like conceptual intelligence.

q-bio.NC↗
arXiv2025

RAJ-PGA: Reasoning-Activated Jailbreak and Principle-Guided Alignment Framework for Large Reasoning Models

Large Reasoning Models (LRMs) face a distinct safety vulnerability: their internal reasoning chains may generate harmful content even when the final output appears benign. To address this overlooked risk, we first propose a novel attack paradigm, Reasoning-Activated Jailbreak (RAJ) via Concretization, which demonstrates that refining malicious prompts to be more specific can trigger step-by-step logical reasoning that overrides the model's safety protocols. To systematically mitigate this vulnerability, we further develop a scalable framework for constructing high-quality safety alignment datasets. This framework first leverages the RAJ attack to elicit challenging harmful reasoning chains from LRMs, then transforms these high-risk traces into safe, constructive, and educational responses through a tailored Principle-Guided Alignment (PGA) mechanism. Then, we introduce the PGA dataset, a verified alignment dataset containing 3,989 samples using our proposed method. Extensive experiments show that fine-tuning LRMs with PGA dataset significantly enhances model safety, achieving up to a 29.5% improvement in defense success rates across multiple jailbreak benchmarks. Critically, our approach not only defends against sophisticated reasoning-based attacks but also preserves, even enhances, the model's general reasoning capabilities. This work provides a scalable and effective pathway for safety alignment in reasoning-intensive AI systems, addressing the core trade-off between safety and functional performance.

cs.AI↗
arXiv2025

Shock Formation for Compressible Euler Equations on $\mathbb{S}^2$

In this paper, we prove the finite-time shock formation for the compressible Euler equations on the two-dimensional sphere $\mathbb{S}^2$. In contrast to the flat Euclidean case $\mathbb{R}^2$, the geometry of $\mathbb S^2$ imposes new difficulties, and the fluid dynamics are affected by the curved background. To overcome these challenges, we modify the existing modulation method and employ a set of carefully constructed, time-dependent coordinates that precisely track the shock formation on $\mathbb{S}^2$. In particular, we first perform a time-dependent rotation of $\mathbb S^2$, then apply the stereographic projection to the sphere, straighten the steepening shock front, and finally construct shock-adapted coordinates. In the shock-adapted coordinates, the compressible Euler equations on $\mathbb{S}^2$ can be recast into a form suitable for self-similar analysis. Within this framework, we implement a detailed bootstrap argument and establish global well-posedness for the self-similar system. After transferring these results back to the original physical system, we thereby demonstrate the finite-time shock formation on $\mathbb{S}^2$.

math.AP↗
arXiv2025

Rethinking Label Consistency of In-Context Learning: An Implicit Transductive Label Propagation Perspective

Large language models (LLMs) perform in-context learning (ICL) with minimal supervised examples, which benefits various natural language processing (NLP) tasks. One of the critical research focus is the selection of prompt demonstrations. Current approaches typically employ retrieval models to select the top-K most semantically similar examples as demonstrations. However, we argue that existing methods are limited since the label consistency is not guaranteed during demonstration selection. Our cognition derives from the Bayesian view of ICL and our rethinking of ICL from the transductive label propagation perspective. We treat ICL as a transductive learning method and incorporate latent concepts from Bayesian view and deduce that similar demonstrations guide the concepts of query, with consistent labels serving as estimates. Based on this understanding, we establish a label propagation framework to link label consistency with propagation error bounds. To model label consistency, we propose a data synthesis method, leveraging both semantic and label information, and use TopK sampling with Synthetic Data (TopK-SD) to acquire demonstrations with consistent labels. TopK-SD outperforms original TopK sampling on multiple benchmarks. Our work provides a new perspective for understanding the working mechanisms within ICL.

cs.AI↗
arXiv2025

Diagnosing and Addressing Pitfalls in KG-RAG Datasets: Toward More Reliable Benchmarking

Knowledge Graph Question Answering (KGQA) systems rely on high-quality benchmarks to evaluate complex multi-hop reasoning. However, despite their widespread use, popular datasets such as WebQSP and CWQ suffer from critical quality issues, including inaccurate or incomplete ground-truth annotations, poorly constructed questions that are ambiguous, trivial, or unanswerable, and outdated or inconsistent knowledge. Through a manual audit of 16 popular KGQA datasets, including WebQSP and CWQ, we find that the average factual correctness rate is only 57 %. To address these issues, we introduce KGQAGen, an LLM-in-the-loop framework that systematically resolves these pitfalls. KGQAGen combines structured knowledge grounding, LLM-guided generation, and symbolic verification to produce challenging and verifiable QA instances. Using KGQAGen, we construct KGQAGen-10k, a ten-thousand scale benchmark grounded in Wikidata, and evaluate a diverse set of KG-RAG models. Experimental results demonstrate that even state-of-the-art systems struggle on this benchmark, highlighting its ability to expose limitations of existing models. Our findings advocate for more rigorous benchmark construction and position KGQAGen as a scalable framework for advancing KGQA evaluation.

cs.CL↗
arXiv2025

Scale-free Characteristics of Multilingual Legal Texts and the Limitations of LLMs

We present a comparative analysis of text complexity across domains using scale-free metrics. We quantify linguistic complexity via Heaps' exponent $β$ (vocabulary growth), Taylor's exponent $α$ (word-frequency fluctuation scaling), compression rate $r$ (redundancy), and entropy. Our corpora span three domains: legal documents (statutes, cases, deeds) as a specialized domain, general natural language texts (literature, Wikipedia), and AI-generated (GPT) text. We find that legal texts exhibit slower vocabulary growth (lower $β$) and higher term consistency (higher $α$) than general texts. Within legal domain, statutory codes have the lowest $β$ and highest $α$, reflecting strict drafting conventions, while cases and deeds show higher $β$ and lower $α$. In contrast, GPT-generated text shows the statistics more aligning with general language patterns. These results demonstrate that legal texts exhibit domain-specific structures and complexities, which current generative models do not fully replicate.

cs.CL↗
arXiv2025

Subjective Camera 1.0: Bridging Human Cognition and Visual Reconstruction through Sequence-Aware Sketch-Guided Diffusion

We introduce the concept of a subjective camera to reconstruct meaningful moments that physical cameras fail to capture. We propose Subjective Camera 1.0, a framework for reconstructing real-world scenes from readily accessible subjective readouts, i.e., textual descriptions and progressively drawn rough sketches. Built on optimization-based alignment of diffusion models, our approach avoids large-scale paired training data and mitigates generalization issues. To address the challenge of integrating multiple abstract concepts in real-world scenarios, we design a Sequence-Aware Sketch-Guided Diffusion framework with three loss terms for concept-wise sequential optimization, following the natural order of subjective readouts. Experiments on two datasets demonstrate that our method achieves state-of-the-art performance in image quality as well as spatial and semantic alignment with target scenes. User studies with 40 participants further confirm that our approach is consistently preferred. Our project page is at: subjective-camera.github.io

cs.CV↗
arXiv2025

Language modulates vision: Evidence from neural networks and human brain-lesion models

Comparing information structures in between deep neural networks (DNNs) and the human brain has become a key method for exploring their similarities and differences. Recent research has shown better alignment of vision-language DNN models, such as CLIP, with the activity of the human ventral occipitotemporal cortex (VOTC) than earlier vision models, supporting the idea that language modulates human visual perception. However, interpreting the results from such comparisons is inherently limited due to the "black box" nature of DNNs. To address this, we combined model-brain fitness analyses with human brain lesion data to examine how disrupting the communication pathway between the visual and language systems causally affects the ability of vision-language DNNs to explain the activity of the VOTC. Across four diverse datasets, CLIP consistently captured unique variance in VOTC neural representations, relative to both label-supervised (ResNet) and unsupervised (MoCo) models. This advantage tended to be left-lateralized at the group level, aligning with the human language network. Analyses of 33 stroke patients revealed that reduced white matter integrity between the VOTC and the language region in the left angular gyrus was correlated with decreased CLIP-brain correspondence and increased MoCo-brain correspondence, indicating a dynamic influence of language processing on the activity of the VOTC. These findings support the integration of language modulation in neurocognitive models of human vision, reinforcing concepts from vision-language DNN models. The sensitivity of model-brain similarity to specific brain lesions demonstrates that leveraging manipulation of the human brain is a promising framework for evaluating and developing brain-like computer models.

q-bio.NC↗
arXiv2025

Open Set Domain Adaptation with Vision-language models via Gradient-aware Separation

Open-Set Domain Adaptation (OSDA) confronts the dual challenge of aligning known-class distributions across domains while identifying target-domain-specific unknown categories. Current approaches often fail to leverage semantic relationships between modalities and struggle with error accumulation in unknown sample detection. We propose to harness Contrastive Language-Image Pretraining (CLIP) to address these limitations through two key innovations: 1) Prompt-driven cross-domain alignment: Learnable textual prompts conditioned on domain discrepancy metrics dynamically adapt CLIP's text encoder, enabling semantic consistency between source and target domains without explicit unknown-class supervision. 2) Gradient-aware open-set separation: A gradient analysis module quantifies domain shift by comparing the L2-norm of gradients from the learned prompts, where known/unknown samples exhibit statistically distinct gradient behaviors. Evaluations on Office-Home show that our method consistently outperforms CLIP baseline and standard baseline. Ablation studies confirm the gradient norm's critical role.

cs.LG↗
arXiv2025

Leader and Follower: Interactive Motion Generation under Trajectory Constraints

With the rapid advancement of game and film production, generating interactive motion from texts has garnered significant attention due to its potential to revolutionize content creation processes. In many practical applications, there is a need to impose strict constraints on the motion range or trajectory of virtual characters. However, existing methods that rely solely on textual input face substantial challenges in accurately capturing the user's intent, particularly in specifying the desired trajectory. As a result, the generated motions often lack plausibility and accuracy. Moreover, existing trajectory - based methods for customized motion generation rely on retraining for single - actor scenarios, which limits flexibility and adaptability to different datasets, as well as interactivity in two-actor motions. To generate interactive motion following specified trajectories, this paper decouples complex motion into a Leader - Follower dynamic, inspired by role allocation in partner dancing. Based on this framework, this paper explores the motion range refinement process in interactive motion generation and proposes a training-free approach, integrating a Pace Controller and a Kinematic Synchronization Adapter. The framework enhances the ability of existing models to generate motion that adheres to trajectory by controlling the leader's movement and correcting the follower's motion to align with the leader. Experimental results show that the proposed approach, by better leveraging trajectory information, outperforms existing methods in both realism and accuracy.

cs.RO↗
arXiv2024

Low-Regularity Local Well-Posedness for the Elastic Wave System

We study the elastic wave system in three spatial dimensions. For admissible harmonic elastic materials, we prove a desired low-regularity local well-posedness result for the corresponding elastic wave equations. For such materials, we can split the dynamics into the divergence-part and the curl-part, and each part satisfies a distinct coupled quasilinear wave system with respect to different acoustical metrics. Our main result is that the Sobolev norm $H^{3+}$ of the divergence-part (the faster-wave part) and the $H^{4+}$ of the curl-part (the slower-wave part) can be controlled in terms of initial data for short times. We note that the Sobolev norm assumption $H^{3+}$ is optimal for the divergence-part. This marks the first favorable low-regularity local well-posedness result for a wave system with multiple wave speeds.

math.AP↗
arXiv2024

i-Rebalance: Personalized Vehicle Repositioning for Supply Demand Balance

Ride-hailing platforms have been facing the challenge of balancing demand and supply. Existing vehicle reposition techniques often treat drivers as homogeneous agents and relocate them deterministically, assuming compliance with the reposition. In this paper, we consider a more realistic and driver-centric scenario where drivers have unique cruising preferences and can decide whether to take the recommendation or not on their own. We propose i-Rebalance, a personalized vehicle reposition technique with deep reinforcement learning (DRL). i-Rebalance estimates drivers' decisions on accepting reposition recommendations through an on-field user study involving 99 real drivers. To optimize supply-demand balance and enhance preference satisfaction simultaneously, i-Rebalance has a sequential reposition strategy with dual DRL agents: Grid Agent to determine the reposition order of idle vehicles, and Vehicle Agent to provide personalized recommendations to each vehicle in the pre-defined order. This sequential learning strategy facilitates more effective policy training within a smaller action space compared to traditional joint-action methods. Evaluation of real-world trajectory data shows that i-Rebalance improves driver acceptance rate by 38.07% and total driver income by 9.97%.

cs.AI↗
arXiv2024

Enhancing Software Effort Estimation through Reinforcement Learning-based Project Management-Oriented Feature Selection

Purpose: The study aims to investigate the application of the data element market in software project management, focusing on improving effort estimation by addressing challenges faced by traditional methods. Design/methodology/approach: This study proposes a solution based on feature selection, utilizing the data element market and reinforcement learning-based algorithms to enhance the accuracy of software effort estimation. It explores the application of the MARLFS algorithm, customizing improvements to the algorithm and reward function. Findings: This study demonstrates that the proposed approach achieves more precise estimation compared to traditional methods, leveraging feature selection to guide project management in software development. Originality/value: This study contributes to the field by offering a novel approach that combines the data element market, machine learning, and feature selection to improve software effort estimation, addressing limitations of traditional methods and providing insights for future research in project management.

cs.SE↗
arXiv2022

$H^{\frac{11}{4}}(\mathbb{R}^2)$ ill-posedness for 2D Elastic Wave system

In this paper, we prove that for the 2D elastic wave equations, a physical system with multiple wave-speeds, its Cauchy problem fails to be locally well-posed in $H^{\frac{11}{4}}(\mathbb{R}^2)$. The ill-posedness here is driven by instantaneous shock formation. In 2D Smith-Tataru showed that the Cauchy problem for a single quasilinear wave equation is locally well-posed in $H^s$ with $s>\frac{11}{4}$. Hence our $H^{\frac{11}{4}}$ ill-posedness obtained here is a desired result. Our proof relies on combining a geometric method and an algebraic wave-decomposition approach, equipped with detailed analysis of the corresponding hyperbolic system.

math.AP↗
arXiv2022

Quantitative blow-up estimates for spacelike singularities in gravitational-collapse cosmological spacetimes

Under spherical symmetry, with double-null coordinates $(u,v)$, we study the gravitational collapse of the Einstein--scalar field system with a positive cosmological constant. The spacetime singularities arise when area radius $r$ vanishes and they are spacelike. We derive new quantitative estimates, obtain polynomial blow-up rates $O(1/r^N)$ for various quantities, and extend the results in [5] by the first author and Zhang and the arguments in [3] by the first author and Gajic to the cosmological settings. In particular, we sharpen the estimates of $r\partial_u r$ and $r\partial_v r$ in [5] and prove that the spacelike singularities where $r(u,v)=0$ are $C^{1,1/3}$ in $(u,v)$ coordinates. As an application, these estimates also give quantitative blow-up upper bounds of fluid velocity and density for the hard-phase model of the Einstein-Euler system under irrotational assumption. Near the timelike infinity, we also generalize the theorems in [3] by linking the precise blow-up rates of the Kretschmann scalar to the exponential Price's law along the event horizon. In cosmological settings, this further reveals the mass-inflation phenomena along the spacelike singularities for the first time.

gr-qc↗