arXiv subjects

Zihao Sun

Publications and source records attributed to Zihao Sun.

12 recordsLinked to original sources

arXiv2026

LifelongCrossNav: Persistent 3D Semantic Memory for Cross-Floor Multi-Object Navigation

Object-goal navigation has made substantial progress in semantic perception and exploration, yet persistent memory for multi-object navigation and cross-floor navigation are still commonly addressed separately. We present LifelongCrossNav, a framework for sequential multi-object ObjectNav in unknown multi-floor indoor environments. Within each episode, the agent receives an ordered sequence of object-goal queries while continuously maintaining a shared sparse 3D semantic voxel memory. This memory incrementally accumulates geometric structure, traversability states, and vision-language features, allowing subsequent object-goal queries to retrieve previously acquired scene information without rebuilding the map. To support persistent search across floors, LifelongCrossNav combines support-aware 3D traversability mapping, stair-specific perception, and direction-aware stair traversal. A unified navigation policy coordinates same-floor frontier exploration, live and historical point-of-interest retrieval, stair navigation, and target-object search and approach. We further introduce HM3D-MFMON, a benchmark for sequential Multi-Floor Multi-Object Navigation built on HM3D scenes, including a dedicated subset in which completing the full sequence of object-goal subtasks requires at least one floor transition. Experimental results show that LifelongCrossNav consistently outperforms a representative planar persistent semantic-map baseline on HM3D-MFMON, demonstrating that persistent 3D semantic memory and cross-floor traversability modeling effectively support sequential multi-object navigation in multi-floor environments. Project page: https://flageval-baai.github.io/LifelongCrossNavPage.

cs.RO↗

arXiv2025

RouteRAG: Efficient Retrieval-Augmented Generation from Text and Graph via Reinforcement Learning

Retrieval-Augmented Generation (RAG) integrates non-parametric knowledge into Large Language Models (LLMs), typically from unstructured texts and structured graphs. While recent progress has advanced text-based RAG to multi-turn reasoning through Reinforcement Learning (RL), extending these advances to hybrid retrieval introduces additional challenges. Existing graph-based or hybrid systems typically depend on fixed or handcrafted retrieval pipelines, lacking the ability to integrate supplementary evidence as reasoning unfolds. Besides, while graph evidence provides relational structures crucial for multi-hop reasoning, it is substantially more expensive to retrieve. To address these limitations, we introduce \model{}, an RL-based framework that enables LLMs to perform multi-turn and adaptive graph-text hybrid RAG. \model{} jointly optimizes the entire generation process via RL, allowing the model to learn when to reason, what to retrieve from either texts or graphs, and when to produce final answers, all within a unified generation policy. To guide this learning process, we design a two-stage training framework that accounts for both task outcome and retrieval efficiency, enabling the model to exploit hybrid evidence while avoiding unnecessary retrieval overhead. Experimental results across five question answering benchmarks demonstrate that \model{} significantly outperforms existing RAG baselines, highlighting the benefits of end-to-end RL in supporting adaptive and efficient retrieval for complex reasoning.

cs.CL↗

arXiv2025

FGNet: Leveraging Feature-Guided Attention to Refine SAM2 for 3D EM Neuron Segmentation

Accurate segmentation of neural structures in Electron Microscopy (EM) images is paramount for neuroscience. However, this task is challenged by intricate morphologies, low signal-to-noise ratios, and scarce annotations, limiting the accuracy and generalization of existing methods. To address these challenges, we seek to leverage the priors learned by visual foundation models on a vast amount of natural images to better tackle this task. Specifically, we propose a novel framework that can effectively transfer knowledge from Segment Anything 2 (SAM2), which is pre-trained on natural images, to the EM domain. We first use SAM2 to extract powerful, general-purpose features. To bridge the domain gap, we introduce a Feature-Guided Attention module that leverages semantic cues from SAM2 to guide a lightweight encoder, the Fine-Grained Encoder (FGE), in focusing on these challenging regions. Finally, a dual-affinity decoder generates both coarse and refined affinity maps. Experimental results demonstrate that our method achieves performance comparable to state-of-the-art (SOTA) approaches with the SAM2 weights frozen. Upon further fine-tuning on EM data, our method significantly outperforms existing SOTA methods. This study validates that transferring representations pre-trained on natural images, when combined with targeted domain-adaptive guidance, can effectively address the specific challenges in neuron segmentation.

cs.CV↗

arXiv2025

Memorization in Large Language Models in Medicine: Prevalence, Characteristics, and Implications

Large Language Models (LLMs) have demonstrated significant potential in medicine, with many studies adapting them through continued pre-training or fine-tuning on medical data to enhance domain-specific accuracy and safety. However, a key open question remains: to what extent do LLMs memorize medical training data. Memorization can be beneficial when it enables LLMs to retain valuable medical knowledge during domain adaptation. Yet, it also raises concerns. LLMs may inadvertently reproduce sensitive clinical content (e.g., patient-specific details), and excessive memorization may reduce model generalizability, increasing risks of misdiagnosis and making unwarranted recommendations. These risks are further amplified by the generative nature of LLMs, which can not only surface memorized content but also produce overconfident, misleading outputs that may hinder clinical adoption. In this work, we present a study on memorization of LLMs in medicine, assessing its prevalence (how frequently it occurs), characteristics (what is memorized), volume (how much content is memorized), and potential downstream impacts (how memorization may affect medical applications). We systematically analyze common adaptation scenarios: (1) continued pretraining on medical corpora, (2) fine-tuning on standard medical benchmarks, and (3) fine-tuning on real-world clinical data, including over 13,000 unique inpatient records from Yale New Haven Health System. The results demonstrate that memorization is prevalent across all adaptation scenarios and significantly higher than that reported in the general domain. Moreover, memorization has distinct characteristics during continued pre-training and fine-tuning, and it is persistent: up to 87% of content memorized during continued pre-training remains after fine-tuning on new medical tasks.

cs.CL↗

arXiv2025

Interfacial Stability in Tensionless Phase-Separated Quorum-Sensing Systems

Interfacial phenomena of motility-induced phase separation of active particles challenge our conventional understanding of phase coexistence. Despite the ubiquity of nonmechanical communication couplings among real active particles, most works on active interface have concentrated on active Brownian systems with steric interparticle interactions. Here, we study the interfacial behavior of phase-separated active particles interacting solely via quorum-sensing communications using both theory and simulations. Strikingly, we find that the quorum-sensing active system exhibits vanishing mechanical surface tension but nonzero effective capillary surface tension. We further demonstrate that the mechanical equilibrium of the tensionless interface is sustained by polarization force at the interface; while its dynamics is governed by the surface stiffness, which arises from tangential particle flux induced by local interfacial deformation. Our work reveals the fundamental distinction between mechanical and capillary surface tensions in active matter and paves the way for future exploration of active interface phenomena.

cond-mat.soft↗

arXiv2025

Intrinsic pressure as a convenient mechanical framework for dry active matter

The identification of local pressure in active matter systems remains a subject of considerable debate. Through theoretical calculations and extensive simulations of various active systems, we demonstrate that intrinsic pressure (defined in the same way as in passive systems) is an ideal candidate for local pressure of dry active matter, while the self-propelling forces on the active particles are considered as effective external forces originating from the environment. Such a framework is universal and especially convenient for analyzing mechanics of dry active systems, and it recovers the conventional scenario of mechanical equilibrium well-known in passive systems. Thus, our work is of fundamental importance to further explore mechanics and thermodynamics of complex active systems.

cond-mat.soft↗

arXiv2025

WebArXiv: Evaluating Multimodal Agents on Time-Invariant arXiv Tasks

Recent progress in large language models (LLMs) has enabled the development of autonomous web agents capable of navigating and interacting with real websites. However, evaluating such agents remains challenging due to the instability and inconsistency of existing benchmarks, which often rely on dynamic content or oversimplified simulations. In this work, we introduce WebArXiv, a static and time-invariant benchmark comprising 275 web-based tasks grounded in the arXiv platform. WebArXiv ensures reproducible and reliable evaluation by anchoring tasks in fixed web snapshots with deterministic ground truths and standardized action trajectories. Through behavioral analysis, we identify a common failure mode, Rigid History Reflection, where agents over-rely on fixed interaction histories. To address this, we propose a lightweight dynamic reflection mechanism that allows agents to selectively retrieve relevant past steps during decision-making. We evaluate ten state-of-the-art web agents on WebArXiv. Results demonstrate clear performance differences across agents and validate the effectiveness of our proposed reflection strategy.

cs.IR↗

arXiv2025

Active Hyperuniform Networks of Chiral Magnetic Micro-Robotic Spinners

Disorder hyperuniform (DHU) systems possess a hidden long-range order manifested as the complete suppression of normalized large-scale density fluctuations like crystals, which endows them with many unique properties. Here, we demonstrate a new organization mechanism for achieving stable DHU structures in active-particle systems via investigating the self-assembly of robotic spinners with three-fold symmetric magnetic binding sites up to a heretofore experimentally unattained system size, i.e., with $\sim 1000$ robots. The spinners can self-organize into a wide spectrum of actively rotating three-coordinated network structures, among which a set of stable DHU networks robustly emerge. These DHU networks are topological transformations of a honeycomb network by continuously introducing the Stone-Wales defects, which are resulted from the competition between tunable magnetic binding and local twist due to active rotation of the robots. Our results reveal novel mechanisms for emergent DHU states in active systems and achieving novel DHU materials with desirable properties.

cond-mat.soft↗

arXiv2024

Data Contamination Can Cross Language Barriers

The opacity in developing large language models (LLMs) is raising growing concerns about the potential contamination of public benchmarks in the pre-training data. Existing contamination detection methods are typically based on the text overlap between training and evaluation data, which can be too superficial to reflect deeper forms of contamination. In this paper, we first present a cross-lingual form of contamination that inflates LLMs' performance while evading current detection methods, deliberately injected by overfitting LLMs on the translated versions of benchmark test sets. Then, we propose generalization-based approaches to unmask such deeply concealed contamination. Specifically, we examine the LLM's performance change after modifying the original benchmark by replacing the false answer choices with correct ones from other questions. Contaminated models can hardly generalize to such easier situations, where the false choices can be \emph{not even wrong}, as all choices are correct in their memorization. Experimental results demonstrate that cross-lingual contamination can easily fool existing detection methods, but not ours. In addition, we discuss the potential utilization of cross-lingual contamination in interpreting LLMs' working mechanisms and in post-training LLMs for enhanced multilingual capabilities. The code and dataset we use can be obtained from \url{https://github.com/ShangDataLab/Deep-Contam}.

cs.CL↗

arXiv2023

Surface Tension Between Coexisting Phases of Active Brownian Particles

The confliction between the stable interface in phase-separated active Brownian particles and its negative surface tension, obtained mechanically via the active pressure, has sparked considerable debate about the formula of active surface tension. Based on the intrinsic pressure of active system, we here derive a new mechanical expression of active surface tension by calculating the work required to create a differential interface area, while remaining the interfacial profiles of intensive quantities invariant (not considered previously). Our expression yields a significantly positive surface tension that increases with the particle activity, which is further supported by mechanical stability analysis of both steady-state droplet and fluctuating interface. Our work is thus promising to resolve the contradiction related to active surface tension.

cond-mat.soft↗

arXiv2023

PA&DA: Jointly Sampling PAth and DAta for Consistent NAS

Based on the weight-sharing mechanism, one-shot NAS methods train a supernet and then inherit the pre-trained weights to evaluate sub-models, largely reducing the search cost. However, several works have pointed out that the shared weights suffer from different gradient descent directions during training. And we further find that large gradient variance occurs during supernet training, which degrades the supernet ranking consistency. To mitigate this issue, we propose to explicitly minimize the gradient variance of the supernet training by jointly optimizing the sampling distributions of PAth and DAta (PA&DA). We theoretically derive the relationship between the gradient variance and the sampling distributions, and reveal that the optimal sampling probability is proportional to the normalized gradient norm of path and training data. Hence, we use the normalized gradient norm as the importance indicator for path and training data, and adopt an importance sampling strategy for the supernet training. Our method only requires negligible computation cost for optimizing the sampling distributions of path and data, but achieves lower gradient variance during supernet training and better generalization performance for the supernet, resulting in a more consistent NAS. We conduct comprehensive comparisons with other improved approaches in various search spaces. Results show that our method surpasses others with more reliable ranking performance and higher accuracy of searched architectures, showing the effectiveness of our method. Code is available at https://github.com/ShunLu91/PA-DA.

cs.CV↗

arXiv2022

ResiDualGAN: Resize-Residual DualGAN for Cross-Domain Remote Sensing Images Semantic Segmentation

The performance of a semantic segmentation model for remote sensing (RS) images pretrained on an annotated dataset would greatly decrease when testing on another unannotated dataset because of the domain gap. Adversarial generative methods, e.g., DualGAN, are utilized for unpaired image-to-image translation to minimize the pixel-level domain gap, which is one of the common approaches for unsupervised domain adaptation (UDA). However, the existing image translation methods are facing two problems when performing RS images translation: 1) ignoring the scale discrepancy between two RS datasets which greatly affects the accuracy performance of scale-invariant objects, 2) ignoring the characteristic of real-to-real translation of RS images which brings an unstable factor for the training of the models. In this paper, ResiDualGAN is proposed for RS images translation, where an in-network resizer module is used for addressing the scale discrepancy of RS datasets, and a residual connection is used for strengthening the stability of real-to-real images translation and improving the performance in cross-domain semantic segmentation tasks. Combined with an output space adaptation method, the proposed method greatly improves the accuracy performance on common benchmarks, which demonstrates the superiority and reliability of ResiDuanGAN. At the end of the paper, a thorough discussion is also conducted to give a reasonable explanation for the improvement of ResiDualGAN. Our source code is available at https://github.com/miemieyanga/ResiDualGAN-DRDG.

cs.CV↗