arXiv subjects

Ming Zhu

Publications and source records attributed to Ming Zhu.

At least 37 records · Page 2Linked to original sources

arXiv2026

Whisper-AuT: Domain-Adapted Audio Encoder for Efficient Audio-LLM Training

Audio-native large language models (audio-LLMs) commonly use Whisper as their audio encoder. However, Whisper was trained exclusively on speech data, producing weak representations for music and environmental sound. This forces downstream audio-LLMs to compensate through extensive training on large-scale non-speech data. We present Whisper-AuT, a domain-adapted audio encoder obtained by fine-tuning Whisper-large-v3 on a curated mixture of speech (80%), environmental sound (10%), and music (10%) totaling approximately 20M samples. The full encoder-decoder is trained end-to-end with a seq2seq captioning objective; the decoder is then discarded and only the encoder is retained. Linear probe evaluations show that Whisper-AuT achieves +23.0% on ESC-50 (environmental sound), +5.0% on GTZAN (music genre), and +0.7% on Speech Commands (keyword spotting) compared to the original Whisperlarge-v3 encoder. Whisper-AuT is designed as a drop-in replacement for Whisper in audio-LLM architectures, with the goal of reducing downstream training cost by providing stronger initial audio representations for non-speech domains.

cs.SD↗

arXiv2026

Optical identification of the FASHI sources: toward the extended Local Volume

We extracted a list of 662 nearby (within $\sim16$ Mpc) HI-detection sources from the Five-hundred-meter Aperture Spherical radio Telescope (FAST) All Sky HI Survey (FASHI) and made a visual identification of them with optical counterpart. This inspection led to the discovery of 71 new dwarf galaxies. All of them are dwarf irregular galaxies with ongoing star formation. They are characterized by the following median parameters: visual magnitude of $g=17.8$ mag and color $(g-r)=0.29$ mag, HI-flux $S_\mathrm{HI}=718$ mJy km/s, HI-mass $M_\mathrm{HI}=3.7\times10^7$ $M_\odot$, as well as the HI line-width of $W_{50}=37$ km/s.

astro-ph.GA↗

arXiv2026

RealUserSim: Bridging the Reality Gap in Agent Benchmarking via Grounded User Simulation

LLM-based user simulation is the primary mechanism for end-to-end agent evaluation, yet simulated users are poor proxies for real humans: unconstrained LLM defaults produce a Formalism Ceiling (style match rates of 6-8% against real users), while hand-crafted behavioral directives trigger Directive Amplification, where models hyper-interpret instructions into unnatural behavioral extremes that vary dramatically across simulator models. We present RealUserSim, the first user simulation framework grounded in real behavioral data. From 14,000+ authentic human-LLM conversations (WildChat), we extract 7,275 executable behavioral profiles and use them to ground LLM simulators. A fidelity benchmark (PT3) on 600 conversations across 71+ domains with anti-leakage controls shows that grounded simulation raises match rate from 24.2% to 45.3% across five behavioral dimensions. Agent evaluation on TauBench with 6 simulator models and extensive analysis shows that grounded simulation acts as a realistic stress test, surfacing three failure mechanisms invisible to cooperative simulators (mean -3.2% to -3.5% task success degradation), while Directive Amplification in existing benchmarks produces unrealistic behavior that compromises the validity of agent evaluation.

cs.HC↗

arXiv2026

Robust topological BIC nanocavities for upconversion directional emission

Photonic bound states in the continuum (BICs) provide a revolutionary paradigm for boosting light-matter interactions in integrated nanocavity systems. Nevertheless, precise manipulation of open cavity-emitter architectures still faces critical challenges, especially in realizing deterministic directional radiation and suppressing the perturbation of intrinsic cavity modes induced by emitters as local impurities. Conventional investigations on cavity-emitter coupling are predominantly based on ensemble measurements, which inevitably mask the intrinsic physics underlying individual light-matter interactions. Here, we propose a robust strategy to control the upconversion and emission of a single-particle emitter using a topological plasmonic cavity with broken σh mirror symmetry. This structured design enables the transition from symmetry-protected BICs to a multi-BIC regime with finite but ultrahigh confinement, where nontrivial phase evolution and hybridization of transverse electric and magnetic modes open a well-defined far-field radiation channel for directional emission. Leveraging this scheme, we experimentally demonstrate dramatically enhanced radiation intensity from a single point-like emitter, together with uniform and deterministic directional emission, while achieving excellent structural robustness against local perturbations. This work establishes a general framework for engineering coherent directional light emission at the nanoscale, which lays a solid foundation for high-performance chip-scale integrated nanophotonic applications.

physics.optics↗

arXiv2026

HI Gas and Star Formation in Major Galaxy Pairs from the FAST All-Sky HI Survey (FASHI)

Atomic hydrogen (HI) plays a fundamental role in fueling star formation in galaxies. However, the behavior of HI gas in interacting systems, particularly galaxy pairs, remains elusive. In this work, we investigate the HI content of major mergers by cross-matching the extragalactic HI catalog from the FAST All-Sky HI Survey (FASHI) with a previously established sample of isolated galaxy pairs. With the superior sensitivity of FAST, we have constructed the largest sample of major mergers with HI detections, consisting of $440$ galaxy pairs: $364$ spiral-spiral (S+S) and $76$ spiral-elliptical (S+E) systems. We examine the HI gas fraction ($f_{\mathrm{HI}}$), star formation rate (SFR) and HI star formation efficiency ($\mathrm{SFE_{HI}}=\mathrm{SFR}/M_{\rm HI}$) for individual galaxies in pairs. The control sample is matched in both stellar mass and redshift. We find that paired galaxies, particularly those in pairs with small projected separations ($d_{\mathrm{p}}<50\ h^{-1}\mathrm{kpc}$), exhibit systematically lower (by $8.8\%$) HI gas fractions compared to the control galaxies. The SFR is enhanced for galaxies in S+S pairs. $\mathrm{SFE_{HI}}$ is $\sim15\%$ higher for galaxies in S+S pairs than in the control galaxies, while spiral galaxies in S+E pairs show no significant difference in $\mathrm{SFE_{HI}}$ compared to the control sample. These findings suggest that the merging process triggers efficient HI gas depletion and enhances star formation, especially in close S+S pairs. Notably, our sample includes $26$ red spirals in paired systems. These galaxies exhibit HI deficiency and suppressed star formation activity compared to the isolated galaxies, indicating that interactions may affect quiescent spirals differently, potentially due to mechanisms similar to ellipticals.

astro-ph.GA↗

arXiv2026

Enterprise Sales Copilot: Enabling Real-Time AI Support with Automatic Information Retrieval in Live Sales Calls

During live sales calls, customers frequently ask detailed product questions that require representatives to manually search internal databases and CRM systems. This process typically takes 25-65 seconds per query, creating awkward pauses that hurt customer experience and reduce sales efficiency. We present SalesCopilot, a real-time AI-powered assistant that eliminates this bottleneck by automatically detecting customer questions, retrieving relevant information from the product database, and displaying concise answers on the representative's dashboard in seconds. The system integrates streaming speech-to-text transcription, large language model (LLM)-based question detection, and retrieval-augmented generation (RAG) over a structured product database into a unified real-time pipeline. We demonstrate SalesCopilot on an insurance sales scenario with 50 products spanning 10 categories (2,490 FAQs, 290 coverage details, and 162 pricing tiers). In our benchmark evaluation, SalesCopilot achieves a measured mean response time of 2.8 seconds with 100% question detection rate, representing a 14xspeedup compared to manual CRM search in an internal study. The system is domain-agnostic and can be adapted to any enterprise sales domain by replacing the product database.

cs.SD↗

arXiv2026

Building Enterprise Realtime Voice Agents from Scratch: A Technical Tutorial

We present a technical tutorial for building enterprise-grade realtime voice agents from first principles. While end-to-end speech-to-speech models may ultimately provide the best latency for voice agents, fully self-hosted end-to-end solutions are not yet available. We evaluate the closest candidate, Qwen3-Omni, across three configurations: its cloud-only DashScope Realtime API achieves $\sim$702ms audio-to-audio latency with streaming, but is not self-hostable; its local vLLM deployment supports only the Thinker (text generation from audio, 516ms), not the Talker (audio synthesis); and its local Transformers deployment runs the full pipeline but at $\sim$146s -- far too slow for realtime. The cascaded streaming pipeline (STT $\rightarrow$ LLM $\rightarrow$ TTS) therefore remains the practical architecture for self-hosted realtime voice agents, and the focus of this tutorial. We build a complete voice agent using Deepgram (streaming STT), vLLM-served LLMs with function calling (streaming text generation), and ElevenLabs (streaming TTS), achieving a measured time-to-first-audio of 755ms (best case 729ms) with full function calling support. We release the full codebase as a 9-chapter progressive tutorial with working, tested code for every component.

cs.SD↗

arXiv2026

\HI 21-cm Line Properties of the Nearby LIRG IRAS 04296+2923

We present an analysis of archival Very Large Array (VLA) and Five-hundred-meter Aperture Spherical radio Telescope (FAST) \HI\ 21 cm data, together with archival multi-band radio continuum observations, of the nearby luminous infrared galaxy IRAS~04296+2923. The system, located behind the Taurus dark cloud at a distance of $\sim$29 Mpc, forms a small galaxy group consisting of five members as revealed by the \HI\ imaging. IRAS~04296+2923 has a close companion, HI~0432+2926, with a projected separation of $\sim$40 kpc, a small line-of-sight velocity difference of $Δ$ v = 26 km s$^{-1}$, and comparable total \HI\ masses of order $10^{9}$~$M_{\odot}$. Both galaxies exhibit regular \HI\ velocity fields and characteristic double-horn profiles in the VLA and FAST data, accompanied by only subtle asymmetries and extended \HI\ structures, indicating rotation-dominated kinematics with early signs of weak tidal interaction. Radio continuum emission is detected only from IRAS~04296+2923 and is confined to its nuclear region, consistent with previous studies. Modeling of its multi-band radio spectrum reveals a significant contribution from free--free emission at high frequencies ($>$30 GHz) and a high FIR-to-radio flux ratio ($q_{8.4}\simeq3.2$), implying a young, dust-obscured nuclear starburst. Taken together, the regular \HI\ kinematics, the small velocity offset, and the group-scale environment favor an interpretation in which IRAS~04296+2923 and HI~0432+2926 form a gravitationally bound, orbiting galaxy pair embedded in a small group, rather than an advanced merger. In this context, the luminous infrared galaxy (LIRG) nature of IRAS~04296+2923 is more plausibly driven by internal processes, such as bar-induced gas inflow, possibly modulated by long-timescale, low-level tidal interactions with nearby group companions.

astro-ph.GA↗

arXiv2026

Position: Vector Prompt Interfaces Should Be Exposed to Enable Customization of Large Language Models

As large language models (LLMs) transition from research prototypes to real-world systems, customization has emerged as a central bottleneck. While text prompts can already customize LLM behavior, we argue that text-only prompting does not constitute a suitable control interface for scalable, stable, and inference-only customization. This position paper argues that model providers should expose \emph{vector prompt inputs} as part of the public interface for customizing LLMs. We support this position with diagnostic evidence showing that vector prompt tuning continues to improve with increasing supervision whereas text-based prompt optimization saturates early, and that vector prompts exhibit dense, global attention patterns indicative of a distinct control mechanism. We further discuss why inference-only customization is increasingly important under realistic deployment constraints, and why exposing vector prompts need not fundamentally increase model leakage risk under a standard black-box threat model. We conclude with a call to action for the community to rethink prompt interfaces as a core component of LLM customization.

cs.CL↗

arXiv2026

VoiceAgentRAG: Solving the RAG Latency Bottleneck in Real-Time Voice Agents Using Dual-Agent Architectures

We present VoiceAgentRAG, an open-source dual-agent memory router that decouples retrieval from response generation. A background Slow Thinker agent continuously monitors the conversation stream, predicts likely follow-up topics using an LLM, and pre-fetches relevant document chunks into a FAISS-backed semantic cache. A foreground Fast Talker agent reads only from this sub-millisecond cache, bypassing the vector database entirely on cache hits.

cs.SD↗

arXiv2026

AudioCapBench: Quick Evaluation on Audio Captioning across Sound, Music, and Speech

We introduce AudioCapBench, a benchmark for evaluating audio captioning capabilities of large multimodal models. \method covers three distinct audio domains, including environmental sound, music, and speech, with 1,000 curated evaluation samples drawn from established datasets. We evaluate 13 models across two providers (OpenAI, Google Gemini) using both reference-based metrics (METEOR, BLEU, ROUGE-L) and an LLM-as-Judge framework that scores predictions on three orthogonal dimensions: \textit{accuracy} (semantic correctness), \textit{completeness} (coverage of reference content), and \textit{hallucination} (absence of fabricated content). Our results reveal that Gemini models generally outperform OpenAI models on overall captioning quality, with Gemini~3~Pro achieving the highest overall score (6.00/10), while OpenAI models exhibit lower hallucination rates. All models perform best on speech captioning and worst on music captioning. We release the benchmark as well as evaluation code to facilitate reproducible audio understanding research.

cs.SD↗

arXiv2026

Prompt Optimization Via Diffusion Language Models

We propose a diffusion-based framework for prompt optimization that leverages Diffusion Language Models (DLMs) to iteratively refine system prompts through masked denoising. By conditioning on interaction traces, including user queries, model responses, and optional feedback, our method enables flexible, span-level prompt updates without requiring gradient access or modifying the downstream language model. Across diverse benchmarks (e.g., $τ$-bench, SST-2, SST-5), DLM-optimized prompts consistently improve the performance of a frozen target LLM (e.g., GPT-4o-mini). We further show that moderate diffusion step counts provide the best balance between refinement quality and stability. These results highlight diffusion-based prompt optimization as a general, model-agnostic, and scalable approach for enhancing LLM performance through iterative prompt refinement.

cs.CL↗

arXiv2026

Visual Attention Reasoning via Hierarchical Search and Self-Verification

Multimodal Large Language Models (MLLMs) frequently hallucinate due to their reliance on fragile, linear reasoning and weak visual grounding. We propose Visual Attention Reasoning (VAR), a reinforcement learning framework that reformulates reasoning as a hierarchical search with self-verification. VAR enforces traceable evidence grounding by generating explicit bounding boxes, guided by a novel reward function combining geometric precision and semantic sufficiency. Furthermore, it replaces linear Chain-of-Thought with a tree-search policy capable of backtracking to correct logical errors. Theoretical analysis validates the framework's reliability, and extensive experiments demonstrate that VAR significantly outperforms state-of-the-art methods on complex hallucination and safety benchmarks.

cs.AI↗

arXiv2026

Unveiling a Thin Filament of the Cosmic Web in the Ursa Major Supergroup

Filaments are crucial components of the cosmic web, representing the extensive and aligned distributions of galaxies and gas. Using the Five-hundred-meter Aperture Spherical radio Telescope (FAST), we report the detection of a filament in the Ursa Major supergroup using atomic-hydrogen (HI) observations. This filament consists of sixteen various types of galaxies and five starless gas clumps, spanning a length of approximately 0.9 Mpc. Notably, it is extremely thin, with a thickness comparable to the diameter of a galaxy. We observed a galaxy-filament spin alignment and a velocity gradient within the filament. These findings strongly suggest a cold accretion flow along the filament, potentially contributing to the formation and growth of the galaxies. The thin filament, as a small group, is likely to be merged into the Ursa Major supergroup in the context of hierarchical structure formation.

astro-ph.CO↗

arXiv2026

Not so-dark: High resolution HI imaging of J0139+4328 and identification of an optical counterpart

Dark galaxies - systems rich in neutral hydrogen (HI) gas but with no stars - are a common prediction of numerous theoretical models and cosmological simulations. However, the unequivocal identification of such sources in current HI surveys has proven challenging. In this work, we present interferometric follow-up observations with the VLA of a former dark galaxy candidate J0139+4328, originally detected with the single-dish FAST telescope. The improved spatial resolution of the VLA data allow us to identify a faint optical counterpart and characterize the galaxy. Located at a distance of about 31 Mpc, J0139+4328 has a stellar mass of 3 x 10^6 M_Sun and a relatively high gas richness of M_HI/M_star = 18. Despite its high ratio, the galaxy is consistent, within the scatter, with the stellar-to-HI mass relation of HI-selected samples in the literature and with the baryonic Tully-Fisher relation (BTFR), although its kinematic measurement is subject to large uncertainties. This case highlights the potential of modern high-sensitivity HI surveys for detecting low surface brightness, gas-rich galaxies, but underscores the need for careful interpretation of low-resolution HI data, with potentially large centroid errors, and for sufficiently deep optical imaging to ensure robust identification.

astro-ph.GA↗

arXiv2025

HI Content of Group Galaxies from the FAST All Sky HI Survey

We investigate the atomic gas (HI) content of galaxies in groups using early data from the FAST All Sky HI survey (FASHI). Taking advantage of FAST's blind, wide-area coverage and uniform sensitivity, we assemble a sample of $230$ group galaxies belonging to $182$ groups at $z\leq0.03$. These groups were identified using a halo-based group finder, and they have an median membership of $4$ galaxies. We also derived a matched control sample of isolated systems, and apply censored-data modeling to include both detections and non-detections. At fixed stellar mass and color, we find that the global median HI fraction of group galaxies differs from that of controls by only $-0.04$ dex ($95\%$ CI [$-0.18,\ 0.16$]), indicating at most a mild average offset. The signal is not uniform across populations: satellites are HI-poor (median $Δf_{\mathrm{HI}}=-0.12$ dex), whereas centrals are not HI-deficient (median $Δf_{\mathrm{HI}}=0.13$ dex). Group galaxies located within $0.5R_{180}$ and in denser systems (richness $>10$ or local density $Σ>10\ \mathrm{gal\ Mpc^{-2}}$) show stronger negative offsets, whereas galaxies in the outskirts are statistically indistinguishable from the controls. These results refine earlier reports of global group HI deficiency: with deeper blind data and uniform treatment of upper limits, we show that HI depletion is primarily confined to satellites and compact cores rather than being ubiquitous across groups.

astro-ph.GA↗

arXiv2025

Gas Accretion from a Neighbouring Galaxy Fuels the Low-luminosity AGN in NGC 4278

How a seemingly `dead' host galaxy provides fuel for its active galactic nuclei (AGN) remains an unresolved problem. Using the Five-hundred-meter Aperture Spherical radio Telescope (FAST), we present a new high-sensitivity atomic-hydrogen (HI) observation toward the nearby elliptical galaxy NGC 4278 and its adjacent region. From the observation, we found that external gas accretion from a neighbouring galaxy fuels the low-luminosity AGN in NGC 4278 through tidal interactions. The accreted gas entering NGC 4278 exhibits a rotating gas disk. And the accreted galaxy has been gas-poor and has an HI to stellar mass ratio of about 0.02. Due to the process of gas accretion, it is likely that relativistic jets are generated in the AGN of NGC 4278. The emission of TeV gamma rays in NGC 4278 is likely to be associated with the newly accreted HI gas.

astro-ph.GA↗

arXiv2025

Aetheria: A multimodal interpretable content safety framework based on multi-agent debate and collaboration

The exponential growth of digital content presents significant challenges for content safety. Current moderation systems, often based on single models or fixed pipelines, exhibit limitations in identifying implicit risks and providing interpretable judgment processes. To address these issues, we propose Aetheria, a multimodal interpretable content safety framework based on multi-agent debate and collaboration.Employing a collaborative architecture of five core agents, Aetheria conducts in-depth analysis and adjudication of multimodal content through a dynamic, mutually persuasive debate mechanism, which is grounded by RAG-based knowledge retrieval.Comprehensive experiments on our proposed benchmark (AIR-Bench) validate that Aetheria not only generates detailed and traceable audit reports but also demonstrates significant advantages over baselines in overall content safety accuracy, especially in the identification of implicit risks. This framework establishes a transparent and interpretable paradigm, significantly advancing the field of trustworthy AI content moderation.

cs.AI↗