arXiv subjects

Ke Wang

Publications and source records attributed to Ke Wang.

At least 235 records · Page 13Linked to original sources

arXiv2025

An ALMA Study of Molecular Complexity in the Hot Core G336.99-00.03 MM1

High-mass star formation involves complex processes, with the hot core phase playing a crucial role in chemical enrichment and the formation of complex organic molecules. However, molecular inventories in hot cores remain limited. Using data from the ALMA Three-millimeter Observations of Massive Star-forming regions survey (ATOMS), the molecular composition and evolutionary stages of two distinct millimeter continuum sources in the high-mass star forming region G336.99-00.03 have been characterized. MM1, with 19 distinct molecular species detected, along with 8 isotopologues and several vibrationally/torsionally excited states, has been identified as a hot core. MM2 with only 5 species identified, was defined as a HII region. Isotopic ratios in MM1 were derived, with $^{12}$C/$^{13}$C ranging from 16.0 to 29.2, $^{16}$O/$^{18}$O at 47.7, and $^{32}$S/$^{34}$S at 19.2. Molecular abundances in MM1 show strong agreement with other sources and three-phase warm-up chemical models within an order of magnitude for most species. Formation pathways of key molecules were explored, revealing chemical links and reaction networks. This study provides a detailed molecular inventory of two millimeter continuum sources, shedding light on the chemical diversity and evolutionary processes in high-mass star-forming regions. The derived molecular parameters and isotopic ratios offer benchmarks for astrochemical models, paving the way for further investigation into the formation and evolution of complex organic molecules during the hot core phase.

astro-ph.GA↗

arXiv2025

ASMA-Tune: Unlocking LLMs' Assembly Code Comprehension via Structural-Semantic Instruction Tuning

Assembly code analysis and comprehension play critical roles in applications like reverse engineering, yet they face substantial challenges due to low information density and a lack of explicit syntactic structures. While traditional masked language modeling (MLM) approaches do not explicitly focus on natural language interaction, emerging decoder-focused large language models (LLMs) demonstrate partial success in binary analysis yet remain underexplored for holistic comprehension. We present Assembly Augmented Tuning, an end-to-end structural-semantic instruction tuning framework that synergizes encoder architecture with decoder-based LLMs through a projector module, where the assembly encoder extracts hardware-level structural features, the projector bridges representations with the semantic space, and the instruction-tuned LLM preserves natural language capabilities. Experimental results demonstrate three key advantages: (1) State-of-the-art performance in assembly comprehension with +39.7% Recall@1 and +17.8% MRR improvements over GPT-4-Turbo, (2) Consistent enhancements across base models (24.6-107.4% Recall@1 and 15.2-106.3% MRR on Qwen2.5-Coder, Deepseek-Coder and CodeLlama variants), and (3) Superior instruction-following capabilities (41.5%-118% improvements) with controlled code generation degradation (-8.9% to -35% across architectures).

cs.SE↗

arXiv2025

SG-Reg: Generalizable and Efficient Scene Graph Registration

This paper addresses the challenges of registering two rigid semantic scene graphs, an essential capability when an autonomous agent needs to register its map against a remote agent, or against a prior map. The hand-crafted descriptors in classical semantic-aided registration, or the ground-truth annotation reliance in learning-based scene graph registration, impede their application in practical real-world environments. To address the challenges, we design a scene graph network to encode multiple modalities of semantic nodes: open-set semantic feature, local topology with spatial awareness, and shape feature. These modalities are fused to create compact semantic node features. The matching layers then search for correspondences in a coarse-to-fine manner. In the back-end, we employ a robust pose estimator to decide transformation according to the correspondences. We manage to maintain a sparse and hierarchical scene representation. Our approach demands fewer GPU resources and fewer communication bandwidth in multi-agent tasks. Moreover, we design a new data generation approach using vision foundation models and a semantic mapping module to reconstruct semantic scene graphs. It differs significantly from previous works, which rely on ground-truth semantic annotations to generate data. We validate our method in a two-agent SLAM benchmark. It significantly outperforms the hand-crafted baseline in terms of registration success rate. Compared to visual loop closure networks, our method achieves a slightly higher registration recall while requiring only 52 KB of communication bandwidth for each query frame. Code available at: \href{http://github.com/HKUST-Aerial-Robotics/SG-Reg}{http://github.com/HKUST-Aerial-Robotics/SG-Reg}.

cs.RO↗

arXiv2025

Unconventional domain tessellations in moiré-of-moiré lattices

Imposing incommensurable periodicity on the periodic atomic lattice can lead to complex structural phases consisting of locally periodic structure bounded by topological defects. Twisted trilayer graphene (TTG) is an ideal material platform to study the interplay between different atomic periodicities, which can be tuned by twist angles between the layers, leading to moiré-of-moiré lattices. Interlayer and intralayer interactions between two interfaces in TTG transform this moiré-of-moiré lattice into an intricate network of domain structures at small twist angles, which can harbor exotic electronic behaviors. Here we report a complete structural phase diagram of TTG with atomic scale lattice reconstruction. Using transmission electron microscopy combined with a new interatomic potential simulation, we show several large-scale moiré lattices, including triangular, kagome, and a corner-shared hexagram-shaped domain pattern. Each domain is bounded by a two-dimensional network of domain wall lattices. In the limit of small twist angles, two competing structural orders-rhombohedral and Bernal stackings-with a slight energy difference, cause unconventional lattice reconstruction with spontaneous symmetry breaking and nematic instability, highlighting the importance of long-range interlayer interactions across entire van der Waals layers. The diverse tessellation of distinct domains, whose topological network can be tuned by the adjustment of the twist angles, establishes TTG as a platform for exploring the interplay between emerging quantum properties and controllable nontrivial lattices.

cond-mat.mtrl-sci↗

arXiv2025

SAKR: Enhancing Retrieval-Augmented Generation via Streaming Algorithm and K-Means Clustering

Retrieval-augmented generation (RAG) has achieved significant success in information retrieval to assist large language models LLMs because it builds an external knowledge database. However, it also has many problems, it consumes a lot of memory because of the enormous database, and it cannot update the established index database in time when confronted with massive streaming data. To reduce the memory required for building the database and maintain accuracy simultaneously, we proposed a new approach integrating a streaming algorithm with k-means clustering into RAG. Our approach applied a streaming algorithm to update the index dynamically and reduce memory consumption. Additionally, the k-means algorithm clusters highly similar documents, and the query time would be shortened. We conducted comparative experiments on four methods, and the results indicated that RAG with streaming algorithm and k-means clusters outperforms traditional RAG in accuracy and memory, particularly when dealing with large-scale streaming data.

cs.IR↗

arXiv2025

MathCoder-VL: Bridging Vision and Code for Enhanced Multimodal Mathematical Reasoning

Natural language image-caption datasets, widely used for training Large Multimodal Models, mainly focus on natural scenarios and overlook the intricate details of mathematical figures that are critical for problem-solving, hindering the advancement of current LMMs in multimodal mathematical reasoning. To this end, we propose leveraging code as supervision for cross-modal alignment, since code inherently encodes all information needed to generate corresponding figures, establishing a precise connection between the two modalities. Specifically, we co-develop our image-to-code model and dataset with model-in-the-loop approach, resulting in an image-to-code model, FigCodifier and ImgCode-8.6M dataset, the largest image-code dataset to date. Furthermore, we utilize FigCodifier to synthesize novel mathematical figures and then construct MM-MathInstruct-3M, a high-quality multimodal math instruction fine-tuning dataset. Finally, we present MathCoder-VL, trained with ImgCode-8.6M for cross-modal alignment and subsequently fine-tuned on MM-MathInstruct-3M for multimodal math problem solving. Our model achieves a new open-source SOTA across all six metrics. Notably, it surpasses GPT-4o and Claude 3.5 Sonnet in the geometry problem-solving subset of MathVista, achieving improvements of 8.9% and 9.2%. The dataset and models will be released at https://github.com/mathllm/MathCoder.

cs.CV↗

arXiv2025

Demonstration of low-overhead quantum error correction codes

Quantum computers hold the potential to surpass classical computers in solving complex computational problems. However, the fragility of quantum information and the error-prone nature of quantum operations make building large-scale, fault-tolerant quantum computers a prominent challenge. To combat errors, pioneering experiments have demonstrated a variety of quantum error correction codes. Yet, most of these codes suffer from low encoding efficiency, and their scalability is hindered by prohibitively high resource overheads. Here, we report the demonstration of two low-overhead quantum low-density parity-check (qLDPC) codes, a distance-4 bivariate bicycle code and a distance-3 qLDPC code, on our latest superconducting processor, Kunlun, featuring 32 long-range-coupled transmon qubits. Utilizing a two-dimensional architecture with overlapping long-range couplers, we demonstrate simultaneous measurements of all nonlocal weight-6 stabilizers via the periodic execution of an efficient syndrome extraction circuit. We achieve a logical error rate per logical qubit per cycle of $(8.91 \pm 0.17)\%$ for the distance-4 bivariate bicycle code with four logical qubits and $(7.77 \pm 0.12)\%$ for the distance-3 qLDPC code with six logical qubits. Our results establish the feasibility of implementing various qLDPC codes with long-range coupled superconducting processors, marking a crucial step towards large-scale low-overhead quantum error correction.

quant-ph↗

arXiv2025

Why is the Star Formation Rate Proportional to Dense Gas Mass?

One of the most profound empirical laws of star formation is the Gao-Solomon relation, a linear correlation between the star formation rate (SFR) and the dense molecular gas mass. It is puzzling how the complicated physics in star-formation results in this surprisingly simple proportionality. Using archival Herschel and Atacama Large Millimeter/submillimeter Array Observations, we derived the masses of the most massive cores ($M^{\rm max}_{\rm core}$) and masses of the gravitationally bound gas ($ M_{\rm gas}^{\rm bound}$) in the parent molecular clouds for a sample of low-mass and high-mass star-forming regions. We discovered a significant correlation $\log(M^{\rm max}_{\rm core}/M_{\odot}) = 0.506 \log(M_{\rm gas}^{\rm bound}/M_{\odot})-0.32$. Our discovered $M^{\rm max}_{\rm core}$-$M_{\rm gas}^{\rm bound}$ correlation can be approximately converted to the Gao-Solomon relation if there is (1) a constant 30% efficiency of converting $M^{\rm max}_{\rm core}$ to the mass of the most massive star ($m^{\rm max}_{\rm star}$), and (2) if SFR and $m^{\rm max}_{\rm star}$ are tightly related through $\log({\rm SFR}/(M_{\odot} {\rm yr}^{-1})) = 2.04 \log(m^{\rm max}_{\rm star}/M_{\odot})-5.80$. Intriguingly, both requirements have been suggested by previous theoretical studies (c.f. Yan et al. 2017). Based on this result, we hypothesize that the Gao-Solomon relation is a consequence of combining the following three non-trivial relations (i) SFR vs. $m^{\rm max}_{\rm star}$, (ii) $m^{\rm max}_{\rm star}$ vs. $M^{\rm max}_{\rm core}$, and (iii) $M^{\rm max}_{\rm core}$ vs. $M_{\rm gas}^{\rm bound}$. This finding may open a new possibility to understand the Gao-Solomon relation in an analytic sense.

astro-ph.GA↗

arXiv2025

Machine learning-enabled atomistic insights into phase boundary engineering of solid-solution ferroelectrics

Atomistic control of phase boundaries is crucial for optimizing the functional properties of solid-solution ferroelectrics, yet their microstructural mechanisms remain elusive. Here, we harness machine-learning-driven molecular dynamics to resolve the phase boundary behavior in the KNbO3-KTaO3 (KNTO) system. Our simulations reveal that chemical composition and ordering enable precise modulation of polymorphic phase boundaries (PPBs), offering a versatile pathway for materials engineering. Diffused PPBs and polar nano regions, predicted by our model, highly match with experiments, underscoring the fidelity of the machine-learning atomistic simulation. Crucially, we identify elastic and electrostatic mismatches between ferroelectric KNbO3 and paraelectric KTaO3 as the driving forces behind complex microstructural evolution. This work not only resolves the longstanding microstructural debate but also establishes a generalizable framework for phase boundary engineering toward next-generation high-performance ferroelectrics.

cond-mat.mtrl-sci↗

arXiv2025

Adaptive Markup Language Generation for Contextually-Grounded Visual Document Understanding

Visual Document Understanding has become essential with the increase of text-rich visual content. This field poses significant challenges due to the need for effective integration of visual perception and textual comprehension, particularly across diverse document types with complex layouts. Moreover, existing fine-tuning datasets for this domain often fall short in providing the detailed contextual information for robust understanding, leading to hallucinations and limited comprehension of spatial relationships among visual elements. To address these challenges, we propose an innovative pipeline that utilizes adaptive generation of markup languages, such as Markdown, JSON, HTML, and TiKZ, to build highly structured document representations and deliver contextually-grounded responses. We introduce two fine-grained structured datasets: DocMark-Pile, comprising approximately 3.8M pretraining data pairs for document parsing, and DocMark-Instruct, featuring 624k fine-tuning data annotations for grounded instruction following. Extensive experiments demonstrate that our proposed model significantly outperforms existing state-of-theart MLLMs across a range of visual document understanding benchmarks, facilitating advanced reasoning and comprehension capabilities in complex visual scenarios. Our code and models are released at https://github. com/Euphoria16/DocMark.

cs.CV↗

arXiv2025

The ALMA-ATOMS Survey: Exploring Protostellar Outflows in HC$_3$N

We present the first systematic study of bipolar outflows using HC$_3$N as a tracer in a sample of 146 massive star-forming regions from ALMA-ATOMS survey. Protostellar outflows arise at the initial stage of star formation as a consequence of active accretion. In general, these outflows play a pivotal role in regulating the star formation processes by injecting energetic material in the parent molecular clouds. In such process, lower velocity components of outflows contain a significant portion of the energy. However, extraction of those component is difficult as the corresponding gas is often mixed with that of the ambient cloud. In our sample, we identified 44 bipolar outflows and one explosive outflow in HC$_3$N (J=11--10). The host clumps of these outflows are found to be at different evolutionary stages, suggesting that outflows in HC$_3$N are detectable in different stages of star formation. Also, the non-correlation of HC$_3$N outflows with clump evolutionary stages suggests that HC$_3$N is an unbiased tracer of outflows. Analyses revealed that HC$_3$N performs slightly better in detecting low-velocity components of outflows than traditionally employed tracers like SiO. The derived outflow parameters (i.e outflow mass, momentum, and energy) show moderate correlations with clump mass and luminosity. Our analysis of outflow opening angles and position-velocity diagrams across the outflow lobes show that, HC$_3$N is not only a good tracer of low-velocity outflows, but can also detect high-velocity collimated outflows. Overall, this study indicates that HC$_3$N can be used as a complementary outflow tracer along with the traditionally known outflow tracers, particularly in the detection of the low-velocity components of outflows.

astro-ph.GA↗

arXiv2025

Tunable atomically enhanced moiré Berry curvatures in twisted triple bilayer graphene

We report a twisted triple bilayer graphene platform consisting of three units of Bernal bilayer graphene consecutively twisted at 1.49° and 1.68°. We demonstrate the atomic reconstruction between the two competing moiré superlattices strongly enhances the Berry curvature of each moiré band insulator state, characterized by measured strong nonlocal valley Hall effect that sensitively depends on the inter-moiré competition strength, tunable by manipulating the out-of-plane carrier distribution. Our study sheds light on the microscopic mechanism of atomic and electronic reconstruction in twisted multilayer systems, by systematically investigating transport signatures of moiré Berry curvature and its enhancement from moiré-of-moiré lattice reconstruction. We show that the reconstructed electronic band can be versatilely tuned by electrostatics, providing an approach toward engineering the band structure and its topology for a quantum material platform with designer electrical and optical properties.

cond-mat.mes-hall↗

arXiv2025

Evidence for Core-Core Collision in Barnard 68

The prestellar core Barnard 68 (B68) is a prototypical source to study the initial conditions and chemical processes of star formation. A previous numerical simulation suggested the southeastern bullet is impacting on the main body of B68. In order to obtain more observational evidence, mapping observations of the ground state SO ($1_0-0_1$) emission line at 30 GHz were made with the Effelsberg 100 m telescope. Based on the velocity field and channel maps derived from SO, three velocity components were clearly detected. The velocity field of the main body indicates rotation and is well fitted by a solid-body rotation model. The measured radial velocity difference between the bullet and the main core is about 0.4 km s$^{-1}$, which is almost equal to the velocity obtained by the previous numerical simulation. Therefore, the bullet is most likely impacting onto the rotating main body of B68. A 1D spherical non-LTE Monte-Carlo radiation transfer RATRAN code is performed to derive the radial abundance profile of SO by analyzing the observed velocity-integrated intensity. SO is depleted inside a 60$^{\prime\prime}$ (0.02 pc) radius from the core. The abundance stays constant at 2.0$\times$10$^{-9}$ for radii larger than 60$^{\prime\prime}$ from the center of the main core. The abundance is enhanced at the interface of the bullet and the main core indicating that shock waves were produced by the collision between the bullet and the main core. In conclusion, based on the kinematical and chemical analysis, our observational results support the previously proposed core-core collision scenario in B68.

astro-ph.GA↗

arXiv2025

In Situ Nanometer-Resolution Strain and Orientation Mapping for Gas-Solid Reactions via Precession-Assisted Four-dimensional Scanning Transmission Electron Microscopy

Chemomechanical interactions in gas or liquid environments are crucial for the functionality and longevity of various materials used in sustainable energy technologies, such as rechargeable batteries, water-splitting catalysts, and next-generation nuclear reactors. A comprehensive understanding of nanoscale strain evolution involved in these processes can advance our knowledge of underlying mechanisms and facilitate material design improvements. However, traditional microscopy workflows face challenges due to trade-offs between field of view (FOV), spatial resolution, temporal resolution, and electron beam damage, particularly in gas or liquid environments. Here, we demonstrate in situ nanometer-resolution strain and orientation mapping in a temperature-controlled gas environment with a large FOV. This is achieved by integrating a microelectromechanical system (MEMS)-based closed-cell TEM holder, precession-assisted four-dimensional scanning transmission electron microscopy (4D-STEM), and a direct electron detector (DED). Using the strain evolution during zirconium initial oxidation as a case study, we first outline critical strategies for focused ion beam gas-cell sample preparation and gas-phase TEM workflows to enhance experimental success. We then show that integrating DED with precession electron diffraction and optimizing gas pressure substantially improve the quantity and quality of the detected Bragg peaks in nano-beam electron diffraction patterns, enabling more precise strain measurements. Furthermore, we introduce a practical protocol to pause the reactions, allowing sufficient time for 4D-STEM data collection while ensuring the temporal resolution needed to resolve material dynamics. Our methodology and workflow provide a robust framework for quantitative analysis of chemomechanical evolutions in materials exposed to gas or liquid environments.

cond-mat.mtrl-sci↗

arXiv2025

Single-Input Multi-Output Model Merging: Leveraging Foundation Models for Dense Multi-Task Learning

Model merging is a flexible and computationally tractable approach to merge single-task checkpoints into a multi-task model. Prior work has solely focused on constrained multi-task settings where there is a one-to-one mapping between a sample and a task, overlooking the paradigm where multiple tasks may operate on the same sample, e.g., scene understanding. In this paper, we focus on the multi-task setting with single-input-multiple-outputs (SIMO) and show that it qualitatively differs from the single-input-single-output model merging settings studied in the literature due to the existence of task-specific decoders and diverse loss objectives. We identify that existing model merging methods lead to significant performance degradation, primarily due to representation misalignment between the merged encoder and task-specific decoders. We propose two simple and efficient fixes for the SIMO setting to re-align the feature representation after merging. Compared to joint fine-tuning, our approach is computationally effective and flexible, and sheds light into identifying task relationships in an offline manner. Experiments on NYUv2, Cityscapes, and a subset of the Taskonomy dataset demonstrate: (1) task arithmetic suffices to enable multi-task capabilities; however, the representations generated by the merged encoder has to be re-aligned with the task-specific heads; (2) the proposed architecture rivals traditional multi-task learning in performance but requires fewer samples and training steps by leveraging the existence of task-specific models.

cs.CV↗

arXiv2025

CFIS-YOLO: A Lightweight Multi-Scale Fusion Network for Edge-Deployable Wood Defect Detection

Wood defect detection is critical for ensuring quality control in the wood processing industry. However, current industrial applications face two major challenges: traditional methods are costly, subjective, and labor-intensive, while mainstream deep learning models often struggle to balance detection accuracy and computational efficiency for edge deployment. To address these issues, this study proposes CFIS-YOLO, a lightweight object detection model optimized for edge devices. The model introduces an enhanced C2f structure, a dynamic feature recombination module, and a novel loss function that incorporates auxiliary bounding boxes and angular constraints. These innovations improve multi-scale feature fusion and small object localization while significantly reducing computational overhead. Evaluated on a public wood defect dataset, CFIS-YOLO achieves a mean Average Precision (mAP@0.5) of 77.5\%, outperforming the baseline YOLOv10s by 4 percentage points. On SOPHON BM1684X edge devices, CFIS-YOLO delivers 135 FPS, reduces power consumption to 17.3\% of the original implementation, and incurs only a 0.5 percentage point drop in mAP. These results demonstrate that CFIS-YOLO is a practical and effective solution for real-world wood defect detection in resource-constrained environments.

cs.CV↗

arXiv2025

Stability diagram of layer-polarized quantum Hall states in twisted trilayer graphene

In the twisted trilayer graphene (tTLG) platform, the rich beating patterns between the three graphene layers give rise to a plethora of new length scales and reconstructed electronic bands arising from the emergent moiré and moiré-of-moiré superlattices. The co-existing lattices and superlattices interact and compete with each other to determine the overall transport properties of tTLG, the hierarchy of which can be electrostatically controlled by tuning the out-of-plane charge distribution or layer polarization. In this work, we measure the stability diagram of layer-polarized quantum Hall states in tTLG by systematically mapping out layer-specific Chern numbers in each layer, and intra- and interlayer Chern transitions as a function of displacement field D and total carrier density n. In contrast to twisted bilayer systems, the rich interplay between the three atomic layers gives rise to a complex layer-polarized stability diagram with unconventional transport features that evolve rapidly with electric and magnetic fields. The stability diagram quantitatively characterizes the interlayer screening and charge distribution in tTLG with implication of strong inter-atomic-layer Coulomb coupling. Our work provides comprehensive guidance and insights into predicting and controlling layer-polarization and interlayer transitions in tTLG, and for tuning the individual role and interactions of each participating constituent towards novel material properties.

cond-mat.mes-hall↗

arXiv2025

Dual-band Unified Exploration of Three CMZ Clouds (DUET). Cloud-wide census of continuum sources showing low spectral indices

The Milky Way's Central Molecular Zone (CMZ) is measured to form stars 10 times less efficiently than in the Galactic disk, based on emission from high-mass stars. However, the CMZ's low-mass protostellar population, which accounts for most of the initial stellar mass budget and star formation rate (SFR), is poorly constrained observationally due to limited sensitivity and resolution. We present the Dual-band Unified Exploration of Three CMZ Clouds (DUET) survey, targeting the 20 km/s Cloud, Sgr C, and Dust Ridge cloud e using the Atacama Large Millimeter/submillimeter Array (ALMA) at 1.3 and 3 mm. The mosaicked observations achieve a comparable resolution of 0.2-0.3" (~1600-2500 au) and a sky coverage of 8.3-10.4 square arcmin, respectively. We report 563 continuum sources at 1.3 mm and 330 at 3 mm, respectively, and a dual-band catalog with 450 continuum sources. These sources are marginally resolved at the 2,000 au resolution. We find a cloud-wide deviation (>70%) from commonly-used dust modified blackbody (MBB) models, characterized by either low spectral indices or low brightness temperatures. Three possible explanations for the deviation are discussed. (1) Optically thick Class 0/I Young stellar objects (YSOs) with very small beam filling factors can lead to lower brightness temperatures than what MBB models predict. (2) Large (mm/cm-sized) dust grains have more significant self-scattering, and therefore frequency-dependent albedo could cause lower spectral indices. (3) Free-free emission over 30 uJy can severely contaminate dust emission and cause low spectral indices for mJy sources in our sample, although the needed number of massive protostars (embedded UCHII regions) is infeasibly high for the normal stellar initial mass function. A reliable measurement of the SFR at low protostellar masses will require future work to distinguish between these possible explanations.

astro-ph.GA↗