arXiv subjects

Ming Zhu

Publications and source records attributed to Ming Zhu.

At least 55 records · Page 3Linked to original sources

arXiv2025

HI-detected Dwarf Galaxies in the FASHI Survey: Insights from Single- and Double-Peaked Emission-Line Samples

We present a sample of low HI mass dwarf galaxies ($M_{\rm HI} < 10^8 M_\odot$) detected by The FAST All Sky HI Survey (FASHI) project. Due to the faint and irregular morphology of these galaxies, the default photometry is often inaccurate. Therefore, we utilized The Dark Energy Camera Legacy Survey (DECaLS) data to perform careful photometric measurements, and find that the low HI mass galaxies have similar stellar mass densities to dwarf elliptical (dE) galaxies. Compared to other dwarf galaxy populations, the HI-selected dwarfs exhibit higher stellar mass densities than ultradiffuse galaxies, and similar densities to HI-selected low-surface-brightness galaxies, albeit with lower stellar masses, suggesting a possible evolutionary connection among these populations. By classifying the galaxies according to their HI spectral-line profiles, we show that the double-peaked sources conform closely to the Tully-Fisher relation, whereas the single-peaked sources follow the Faber-Jackson relation but with large scatter. This indicates that the single-peaked systems are likely dispersion dominated and that the relationship between stellar mass and halo mass in such systems may remain consistent across both low- and high-mass regimes. These findings suggest that HI-selected dwarf galaxies with single-peaked HI profiles may share a similar dynamical state with massive ellipticals, offering new insights into their structural evolution and the diversity of formation pathways for low-mass galaxies.

astro-ph.GA↗

arXiv2025

LoCoBench-Agent: An Interactive Benchmark for LLM Agents in Long-Context Software Engineering

As large language models (LLMs) evolve into sophisticated autonomous agents capable of complex software development tasks, evaluating their real-world capabilities becomes critical. While existing benchmarks like LoCoBench~\cite{qiu2025locobench} assess long-context code understanding, they focus on single-turn evaluation and cannot capture the multi-turn interactive nature, tool usage patterns, and adaptive reasoning required by real-world coding agents. We introduce \textbf{LoCoBench-Agent}, a comprehensive evaluation framework specifically designed to assess LLM agents in realistic, long-context software engineering workflows. Our framework extends LoCoBench's 8,000 scenarios into interactive agent environments, enabling systematic evaluation of multi-turn conversations, tool usage efficiency, error recovery, and architectural consistency across extended development sessions. We also introduce an evaluation methodology with 9 metrics across comprehension and efficiency dimensions. Our framework provides agents with 8 specialized tools (file operations, search, code analysis) and evaluates them across context lengths ranging from 10K to 1M tokens, enabling precise assessment of long-context performance. Through systematic evaluation of state-of-the-art models, we reveal several key findings: (1) agents exhibit remarkable long-context robustness; (2) comprehension-efficiency trade-off exists with negative correlation, where thorough exploration increases comprehension but reduces efficiency; and (3) conversation efficiency varies dramatically across models, with strategic tool usage patterns differentiating high-performing agents. As the first long-context LLM agent benchmark for software engineering, LoCoBench-Agent establishes a rigorous foundation for measuring agent capabilities, identifying performance gaps, and advancing autonomous software development at scale.

cs.SE↗

arXiv2025

GeoGNN: Quantifying and Mitigating Semantic Drift in Text-Attributed Graphs

Graph neural networks (GNNs) on text--attributed graphs (TAGs) typically encode node texts using pretrained language models (PLMs) and propagate these embeddings through linear neighborhood aggregation. However, the representation spaces of modern PLMs are highly non--linear and geometrically structured, where textual embeddings reside on curved semantic manifolds rather than flat Euclidean spaces. Linear aggregation on such manifolds inevitably distorts geometry and causes semantic drift--a phenomenon where aggregated representations deviate from the intrinsic manifold, losing semantic fidelity and expressive power. To quantitatively investigate this problem, this work introduces a local PCA--based metric that measures the degree of semantic drift and provides the first quantitative framework to analyze how different aggregation mechanisms affect manifold structure. Building upon these insights, we propose Geodesic Aggregation, a manifold--aware mechanism that aggregates neighbor information along geodesics via log--exp mappings on the unit sphere, ensuring that representations remain faithful to the semantic manifold during message passing. We further develop GeoGNN, a practical instantiation that integrates spherical attention with manifold interpolation. Extensive experiments across four benchmark datasets and multiple text encoders show that GeoGNN substantially mitigates semantic drift and consistently outperforms strong baselines, establishing the importance of manifold--aware aggregation in text--attributed graph learning.

cs.LG↗

arXiv2025

Eleven Local Volume dwarf galaxies in the FASHI survey

We determined HI parameters for eleven nearby late-type dwarf galaxies using FASHI data cubes, despite the fact that the first version of the FASHI catalog does not list any radio sources that could correspond to these galaxies. Four of them are probable peripheral satellites of the bright spiral galaxies: NGC 3556, NGC 4258, NGC 4274, NGC 4490, while others are isolated objects. The considered sample has the following median parameters: a heliocentric velocity of $V_\mathrm{h} = 542 \ km/s$, an HI-line width of $W_\mathrm{50} = 28 \ km/s$, an hydrogen mass of $\log (M_{HI} / M_\odot) = 6.83$, a stellar mass of $\log (M_\star / M_\odot) = 7.19$, and a specific star formation rate of $\mathrm{sSFR} = -10.17 \ yr^{-1}$.

astro-ph.GA↗

arXiv2025

LoCoBench: A Benchmark for Long-Context Large Language Models in Complex Software Engineering

The emergence of long-context language models with context windows extending to millions of tokens has created new opportunities for sophisticated code understanding and software development evaluation. We propose LoCoBench, a comprehensive benchmark specifically designed to evaluate long-context LLMs in realistic, complex software development scenarios. Unlike existing code evaluation benchmarks that focus on single-function completion or short-context tasks, LoCoBench addresses the critical evaluation gap for long-context capabilities that require understanding entire codebases, reasoning across multiple files, and maintaining architectural consistency across large-scale software systems. Our benchmark provides 8,000 evaluation scenarios systematically generated across 10 programming languages, with context lengths spanning 10K to 1M tokens, a 100x variation that enables precise assessment of long-context performance degradation in realistic software development settings. LoCoBench introduces 8 task categories that capture essential long-context capabilities: architectural understanding, cross-file refactoring, multi-session development, bug investigation, feature implementation, code comprehension, integration testing, and security analysis. Through a 5-phase pipeline, we create diverse, high-quality scenarios that challenge LLMs to reason about complex codebases at unprecedented scale. We introduce a comprehensive evaluation framework with 17 metrics across 4 dimensions, including 8 new evaluation metrics, combined in a LoCoBench Score (LCBS). Our evaluation of state-of-the-art long-context models reveals substantial performance gaps, demonstrating that long-context understanding in complex software development represents a significant unsolved challenge that demands more attention. LoCoBench is released at: https://github.com/SalesforceAIResearch/LoCoBench.

cs.SE↗

arXiv2025

Graph-Based Spatio-temporal Attention and Multi-Scale Fusion for Clinically Interpretable, High-Fidelity Fetal ECG Extraction

Congenital Heart Disease (CHD) is the most common neonatal anomaly, highlighting the urgent need for early detection to improve outcomes. Yet, fetal ECG (fECG) signals in abdominal ECG (aECG) are often masked by maternal ECG and noise, challenging conventional methods under low signal-to-noise ratio (SNR) conditions. We propose FetalHealthNet (FHNet), a deep learning framework that integrates Graph Neural Networks with a multi-scale enhanced transformer to dynamically model spatiotemporal inter-lead correlations and extract clean fECG signals. On benchmark aECG datasets, FHNet consistently outperforms long short-term memory (LSTM) models, standard transformers, and state-of-the-art models, achieving R2>0.99 and RMSE = 0.015 even under severe noise. Interpretability analyses highlight physiologically meaningful temporal and lead contributions, supporting model transparency and clinical trust. FHNet illustrates the potential of AI-driven modeling to advance fetal monitoring and enable early CHD screening, underscoring the transformative impact of next-generation biomedical signal processing.

eess.SP↗

arXiv2025

Why Not Transform Chat Large Language Models to Non-English?

The scarcity of non-English data limits the development of non-English large language models (LLMs). Transforming English-centric LLMs to non-English has been identified as an effective and resource-efficient method. Previous works start from base LLMs and perform knowledge distillation (KD) with data generated by stronger LLMs, e.g. GPT-4. Compared to base LLMs, chat LLMs are further optimized for advanced abilities, e.g. multi-turn conversation and human preference alignment, and thus more powerful in both helpfulness and safety. However, transforming a chat LLM involves two critical issues: (1) How can we effectively transfer advanced abilities without their supervised data? (2) How can we prevent the original knowledge from catastrophic forgetting during transformation? We target these issues by introducing a simple framework called TransLLM. For the first issue, TransLLM divides the transfer problem into some common sub-tasks with the translation chain-of-thought, which uses the translation as the bridge between English and non-English step-by-step. We further enhance the performance of sub-tasks with publicly available data. For the second issue, we propose a method comprising two synergistic components: low-rank adaptation for training to maintain the original LLM parameters, and recovery KD, which utilizes data generated by the chat LLM itself to recover the original knowledge from the frozen parameters. In the experiments, we transform the LLaMA-2-chat-7B to the Thai language. Our method, using only single-turn data, outperforms strong baselines and ChatGPT on multi-turn benchmark MT-bench. Furthermore, our method, without safety data, rejects more harmful queries of safety benchmark AdvBench than both ChatGPT and GPT-4. Code is available at https://github.com/hy5468/TransLLM.

cs.CL↗

arXiv2025

PersonaBench: Evaluating AI Models on Understanding Personal Information through Accessing (Synthetic) Private User Data

Personalization is critical in AI assistants, particularly in the context of private AI models that work with individual users. A key scenario in this domain involves enabling AI models to access and interpret a user's private data (e.g., conversation history, user-AI interactions, app usage) to understand personal details such as biographical information, preferences, and social connections. However, due to the sensitive nature of such data, there are no publicly available datasets that allow us to assess an AI model's ability to understand users through direct access to personal information. To address this gap, we introduce a synthetic data generation pipeline that creates diverse, realistic user profiles and private documents simulating human activities. Leveraging this synthetic data, we present PersonaBench, a benchmark designed to evaluate AI models' performance in understanding personal information derived from simulated private user data. We evaluate Retrieval-Augmented Generation (RAG) pipelines using questions directly related to a user's personal information, supported by the relevant private documents provided to the models. Our results reveal that current retrieval-augmented AI models struggle to answer private questions by extracting personal information from user documents, highlighting the need for improved methodologies to enhance personalization capabilities in AI.

cs.AI↗

arXiv2025

Promptomatix: An Automatic Prompt Optimization Framework for Large Language Models

Large Language Models (LLMs) perform best with well-crafted prompts, yet prompt engineering remains manual, inconsistent, and inaccessible to non-experts. We introduce Promptomatix, an automatic prompt optimization framework that transforms natural language task descriptions into high-quality prompts without requiring manual tuning or domain expertise. Promptomatix supports both a lightweight meta-prompt-based optimizer and a DSPy-powered compiler, with modular design enabling future extension to more advanced frameworks. The system analyzes user intent, generates synthetic training data, selects prompting strategies, and refines prompts using cost-aware objectives. Evaluated across 5 task categories, Promptomatix achieves competitive or superior performance compared to existing libraries, while reducing prompt length and computational overhead making prompt optimization scalable and efficient.

cs.CL↗

arXiv2025

APIGen-MT: Agentic Pipeline for Multi-Turn Data Generation via Simulated Agent-Human Interplay

Training effective AI agents for multi-turn interactions requires high-quality data that captures realistic human-agent dynamics, yet such data is scarce and expensive to collect manually. We introduce APIGen-MT, a two-phase framework that generates verifiable and diverse multi-turn agent data. In the first phase, our agentic pipeline produces detailed task blueprints with ground-truth actions, leveraging a committee of LLM reviewers and iterative feedback loops. These blueprints are then transformed into complete interaction trajectories through simulated human-agent interplay. We train a family of models -- the xLAM-2-fc-r series with sizes ranging from 1B to 70B parameters. Our models outperform frontier models such as GPT-4o and Claude 3.5 on $τ$-bench and BFCL benchmarks, with the smaller models surpassing their larger counterparts, particularly in multi-turn settings, while maintaining superior consistency across multiple trials. Comprehensive experiments demonstrate that our verified blueprint-to-details approach yields high-quality training data, enabling the development of more reliable, efficient, and capable agents. We open-source 5K synthetic data trajectories and the trained xLAM-2-fc-r models to advance research in AI agents. Models at https://huggingface.co/collections/Salesforce/xlam-2-67ef5be12949d8dcdae354c4; Dataset at https://huggingface.co/datasets/Salesforce/APIGen-MT-5k and Website at https://apigen-mt.github.io

cs.CL↗

arXiv2025

ActionStudio: A Lightweight Framework for Data and Training of Large Action Models

Large Action models are essential for enabling autonomous agents to perform complex tasks. However, training such models remains challenging due to the diversity of agent environments and the complexity of noisy agentic data. Existing infrastructure offers limited support for scalable, agent-specific fine-tuning and standardized agent data processing. We introduce ActionStudio, a lightweight and extensible data and training framework designed for large action models. ActionStudio unifies diverse agent trajectories using our proposed Unified Format 2.0, supports a range of training workflows with optimized multi-node distributed setup, and integrates robust preprocessing and real-time verification tools. ActionStudio demonstrates up to 9x higher throughput compared to existing agentic training frameworks, and our trained models yield top performances across public and realistic agent benchmarks. To support the broader research community, we open-source the ActionStudio framework and release actionstudio-98k, a curated dataset of 98k high-quality trajectories. Code: https://github.com/SalesforceAIResearch/xLAM.

cs.AI↗

arXiv2025

Deep neutral hydrogen observations of the early-type galaxy NGC 2768: collided by a newly discovered satellite galaxy?

We present the results of a deep neutral hydrogen (Hi) observation of the early-type galaxy NGC 2768 using the Five-hundred-meter Aperture Spherical radio Telescope (FAST). Leveraging the high sensitivity of FAST, we discover an extended gas envelope around NGC 2768. The total Hi mass is measured to be 8.1 x 10^8 M_sun , representing a magnitude increase compared to previous Westerbork Synthesis Radio Telescope (WSRT) studies. Position-velocity (PV) diagram indicates the envelope mainly involves two components: an Hi disk of NGC 2768 and a newly discovered satellite galaxy without detectable counterparts in currently deep optical surveys. The center of the gas disk is mis-aligned with the optical disk of NGC 2768, with more gas redshifted, indicating it has been disturbed. Our study indicates NGC 2768 is currently undergoing a transition from a spiral galaxy to an S0. Previous deep WSRT observations reveal two dense clumps (named as Clumps A and Clump B throughout this paper) in the center of the envelope. We find Clump A corresponds to the densest part of the disk, while Clump B might be a newly discovered satellite galaxy which probably collided NGC 2768 about 0.38 Gyr ago. We also find tidal interactions between Clump B and PGC 2599651, NGC 2768 and UGC 4808. Based on these new findings, we finally analyze hierarchical accretion history of NGC 2768.

astro-ph.GA↗

arXiv2025

ToolScan: A Benchmark for Characterizing Errors in Tool-Use LLMs

Evaluating Large Language Models (LLMs) is one of the most critical aspects of building a performant compound AI system. Since the output from LLMs propagate to downstream steps, identifying LLM errors is crucial to system performance. A common task for LLMs in AI systems is tool use. While there are several benchmark environments for evaluating LLMs on this task, they typically only give a success rate without any explanation of the failure cases. To solve this problem, we introduce TOOLSCAN, a new benchmark to identify error patterns in LLM output on tool-use tasks. Our benchmark data set comprises of queries from diverse environments that can be used to test for the presence of seven newly characterized error patterns. Using TOOLSCAN, we show that even the most prominent LLMs exhibit these error patterns in their outputs. Researchers can use these insights from TOOLSCAN to guide their error mitigation strategies.

cs.SE↗

arXiv2025

LAM SIMULATOR: Advancing Data Generation for Large Action Model Training via Online Exploration and Trajectory Feedback

Large Action Models (LAMs) for AI Agents offer incredible potential but face challenges due to the need for high-quality training data, especially for multi-steps tasks that involve planning, executing tool calls, and responding to feedback. To address these issues, we present LAM SIMULATOR, a comprehensive framework designed for online exploration of agentic tasks with high-quality feedback. Our framework features a dynamic task query generator, an extensive collection of tools, and an interactive environment where Large Language Model (LLM) Agents can call tools and receive real-time feedback. This setup enables LLM Agents to explore and solve tasks autonomously, facilitating the discovery of multiple approaches to tackle any given task. The resulting action trajectory data are then used to create high-quality training datasets for LAMs. Our experiments on popular agentic benchmarks, ToolBench and CRMArena, highlight the effectiveness of LAM SIMULATOR: models trained with self-generated datasets using our framework achieve significant performance gains, up to a 49.3\% improvement over their original baselines. LAM SIMULATOR requires minimal human input during dataset creation, highlighting LAM SIMULATOR's efficiency and effectiveness in speeding up development of AI agents.

cs.CL↗

arXiv2025

The FAST Ursa Major supergroup HI survey (FUMaS): catalog and HI mass function

Using the Five-hundred-meter Aperture Spherical radio Telescope (FAST), we have performed an Ursa Major supergroup HI Survey (FUMaS) covering the entire UMa region centered at RA=11$^h$59$^m$28$^s$.3, DEC=49$^\circ$05'18" with a radius of 7.5$^\circ$. We have obtained the most complete catalog of HI sources in the UMa supergroup, containing 178 HI sources with velocities in the range 625-1213.4 km s$^{-1}$ and masses in the range 10$^{6.0}$-10$^{10.1}$ M$_{\odot}$ assuming a distance of 17.4 Mpc. Among them, 55 HI sources were detected for the first time, of which 32 do not have known optical redshifts. For these 32 sources, we have searched the DESI Legacy Surveys and found optical counterparts for 25 of them (with optical images, but no redshifts), with the remaining 7 sources to be pure HI clouds without an optical counterpart. We detected HI distributions in some interacting systems and discussed four small groups in detail. We computed the HIMF of the UMa supergroup using the 1/V$_\mathrm{max}$ method and fitting it with the non-linear least squares (NLLS) and modified maximum likelihood (MML) methods. We obtained the following HIMF parameters: log$_{10}$($ϕ_*$/Mpc$^{-3}$) = -0.78 $\pm$ 0.15, $α$ = -1.05 $\pm$ 0.07 and log$_{10}$($M_*$/$M_{\odot}$) = 9.87 $\pm$ 0.19 for the NLLS method, and log$_{10}$($ϕ_*$/Mpc$^{-3}$) = -0.70 $\pm$ 0.11, $α$ = -1.05 $\pm$ 0.05 and log$_{10}$($M_*$/$M_{\odot}$) = 9.77 $\pm$ 0.13 for the MML method. This result is similar to that derived from the VLA blind survey, but the slope is steeper because we detected more low-mass galaxies. The slope is flatter than that of the global HIMF, which agrees with the theoretical prediction that galaxies in high-density regions are stripped of gas due to interactions.

astro-ph.GA↗

arXiv2025

FullStack Bench: Evaluating LLMs as Full Stack Coders

As the capabilities of code large language models (LLMs) continue to expand, their applications across diverse code intelligence domains are rapidly increasing. However, most existing datasets only evaluate limited application domains. To address this gap, we have developed a comprehensive code evaluation dataset FullStack Bench focusing on full-stack programming, which encompasses a wide range of application domains (e.g., basic programming, data analysis, software engineering, mathematics, and machine learning). Besides, to assess multilingual programming capabilities, in FullStack Bench, we design real-world instructions and corresponding unit test cases from 16 widely-used programming languages to reflect real-world usage scenarios rather than simple translations. Moreover, we also release an effective code sandbox execution tool (i.e., SandboxFusion) supporting various programming languages and packages to evaluate the performance of our FullStack Bench efficiently. Comprehensive experimental results on our FullStack Bench demonstrate the necessity and effectiveness of our FullStack Bench and SandboxFusion.

cs.AI↗

arXiv2025

Deep Reinforcement Learning for Traffic Light Control in Intelligent Transportation Systems

Smart traffic lights in intelligent transportation systems (ITSs) are envisioned to greatly increase traffic efficiency and reduce congestion. Deep reinforcement learning (DRL) is a promising approach to adaptively control traffic lights based on the real-time traffic situation in a road network. However, conventional methods may suffer from poor scalability. In this paper, we investigate deep reinforcement learning to control traffic lights, and both theoretical analysis and numerical experiments show that the intelligent behavior ``greenwave" (i.e., a vehicle will see a progressive cascade of green lights, and not have to brake at any intersection) emerges naturally a grid road network, which is proved to be the optimal policy in an avenue with multiple cross streets. As a first step, we use two DRL algorithms for the traffic light control problems in two scenarios. In a single road intersection, we verify that the deep Q-network (DQN) algorithm delivers a thresholding policy; and in a grid road network, we adopt the deep deterministic policy gradient (DDPG) algorithm. Secondly, numerical experiments show that the DQN algorithm delivers the optimal control, and the DDPG algorithm with passive observations has the capability to produce on its own a high-level intelligent behavior in a grid road network, namely, the ``greenwave" policy emerges. We also verify the ``greenwave" patterns in a $5 \times 10$ grid road network. Thirdly, the ``greenwave" patterns demonstrate that DRL algorithms produce favorable solutions since the ``greenwave" policy shown in experiment results is proved to be optimal in a specified traffic model (an avenue with multiple cross streets). The delivered policies both in a single road intersection and a grid road network demonstrate the scalability of DRL algorithms.

cs.LG↗

arXiv2025

FAST Reveals the Extended HI Halo and Accretion Signatures of NGC2683

We present the results of our recent HI observations conducted on the edge-on galaxy NGC2683 using the Five-hundred-meter Aperture Spherical radio Telescope (FAST). In comparison to previous observations made by the VLA, FAST has detected a more extensive distribution of HI. Particularly noteworthy is that the detections made by FAST extend approximately four times farther than those of the VLA in the vertical direction from the galactic plane. The total HI flux measured for NGC2683 amounts to $F_{\rm HI} = 112.1\,\rm{Jy\,km\,s^{-1}}$ (equivalent to a total HI mass of $M_{\rm HI} = 2.32 \times 10^9\,{\rm M_\odot}$), which is slightly higher than that detected by VLA. FAST has also identified three dwarf galaxies in close proximity to NGC2683, namely KK69, NGC2683dw1 (hereafter dw1), and NGC2683dw3$?$ (hereafter dw3$?$). dw3$?$ is situated within the extended HI distribution of NGC2683 in projection and lies near the tail of KK69 extending towards NGC2683. These observations suggest that dw3$?$ is likely a result of the accretion process from NGC2683 to KK69. Furthermore, FAST has detected three high-velocity clouds (HVCs), with complex B potentially undergoing accretion with NGC2683. Based on the model from Vollmer et al. 2016 and incorporating the HI halo component, we found that the model with the added HI halo aligns more closely with our FAST observations in NGC2683. The estimated mass of this HI halo is $3 \times 10^8\,{\rm M_\odot}$, constituting approximately 13% of the total HI mass of the galaxy. We suggest that the origination of this HI halo is more likely attributed to external gas accretion.

astro-ph.GA↗