arXiv subjects

Heng Wang

Publications and source records attributed to Heng Wang.

At least 91 records · Page 5Linked to original sources

arXiv2025

ChoreoMuse: Robust Music-to-Dance Video Generation with Style Transfer and Beat-Adherent Motion

Modern artistic productions increasingly demand automated choreography generation that adapts to diverse musical styles and individual dancer characteristics. Existing approaches often fail to produce high-quality dance videos that harmonize with both musical rhythm and user-defined choreography styles, limiting their applicability in real-world creative contexts. To address this gap, we introduce ChoreoMuse, a diffusion-based framework that uses SMPL format parameters and their variation version as intermediaries between music and video generation, thereby overcoming the usual constraints imposed by video resolution. Critically, ChoreoMuse supports style-controllable, high-fidelity dance video generation across diverse musical genres and individual dancer characteristics, including the flexibility to handle any reference individual at any resolution. Our method employs a novel music encoder MotionTune to capture motion cues from audio, ensuring that the generated choreography closely follows the beat and expressive qualities of the input music. To quantitatively evaluate how well the generated dances match both musical and choreographic styles, we introduce two new metrics that measure alignment with the intended stylistic cues. Extensive experiments confirm that ChoreoMuse achieves state-of-the-art performance across multiple dimensions, including video quality, beat alignment, dance diversity, and style adherence, demonstrating its potential as a robust solution for a wide range of creative applications. Video results can be found on our project page: https://choreomuse.github.io.

cs.GR↗

arXiv2025

Step-3 is Large yet Affordable: Model-system Co-design for Cost-effective Decoding

Large language models (LLMs) face low hardware efficiency during decoding, especially for long-context reasoning tasks. This paper introduces Step-3, a 321B-parameter VLM with hardware-aware model-system co-design optimized for minimizing decoding costs. Step-3 innovates in two key dimensions: (1) A novel Multi-Matrix Factorization Attention (MFA) mechanism that significantly reduces both KV cache size and computation while maintaining high attention expressiveness, and (2) Attention-FFN Disaggregation (AFD), a distributed inference system that decouples attention and Feed-Forward Network (FFN) layers into specialized subsystems. This co-design achieves unprecedented cost efficiency: Step-3 significantly reduces theoretical decoding costs compared with models like DeepSeek-V3 and Qwen3 MoE 235B, with the gains widening at longer context. Step-3 achieves low cost while activating 38B parameters per token (more than DeepSeek-V3 and Qwen3 MoE 235B), demonstrating that hardware-aligned attention arithmetic intensity, MoE sparsity, and AFD are critical to cost-effectiveness. We perform a head-to-head comparison with DeepSeek-V3 in its favorable scenarios. Our implementation on Hopper GPUs achieves a decoding throughput of up to 4,039 tokens per second per GPU under 50ms TPOT SLA (4K context, FP8, no MTP). It is higher than DeepSeek-V3's 2,324 in the same setup and sets a new Pareto frontier for LLM decoding.

cs.LG↗

arXiv2025

DesignLab: Designing Slides Through Iterative Detection and Correction

Designing high-quality presentation slides can be challenging for non-experts due to the complexity involved in navigating various design choices. Numerous automated tools can suggest layouts and color schemes, yet often lack the ability to refine their own output, which is a key aspect in real-world workflows. We propose DesignLab, which separates the design process into two roles, the design reviewer, who identifies design-related issues, and the design contributor who corrects them. This decomposition enables an iterative loop where the reviewer continuously detects issues and the contributor corrects them, allowing a draft to be further polished with each iteration, reaching qualities that were unattainable. We fine-tune large language models for these roles and simulate intermediate drafts by introducing controlled perturbations, enabling the design reviewer learn design errors and the contributor learn how to fix them. Our experiments show that DesignLab outperforms existing design-generation methods, including a commercial tool, by embracing the iterative nature of designing which can result in polished, professional slides.

cs.CV↗

arXiv2025

Cosmos World Foundation Model Platform for Physical AI

Physical AI needs to be trained digitally first. It needs a digital twin of itself, the policy model, and a digital twin of the world, the world model. In this paper, we present the Cosmos World Foundation Model Platform to help developers build customized world models for their Physical AI setups. We position a world foundation model as a general-purpose world model that can be fine-tuned into customized world models for downstream applications. Our platform covers a video curation pipeline, pre-trained world foundation models, examples of post-training of pre-trained world foundation models, and video tokenizers. To help Physical AI builders solve the most critical problems of our society, we make Cosmos open-source and our models open-weight with permissive licenses available via https://github.com/nvidia-cosmos/cosmos-predict1.

cs.CV↗

arXiv2025

Nodeless superconducting gap and electron-boson coupling in (La,Pr,Sm)$_{3}$Ni$_2$O$_7$ films

The discovery of superconductivity in Ruddlesden-Popper (RP) bilayer nickelate films under ambient pressure provides an unprecedented opportunity to directly investigate electronic energy scales of the superconducting state and the pairing mechanism. Here, we report angle-resolved photoemission spectroscopy measurements of superconducting (La,Pr,Sm)$_{3}$Ni$_2$O$_7$ thin films epitaxially grown on SrLaAlO$_4$ substrates by developing an ultra-high vacuum low-temperature quenching and transfer technique. A finite superconducting gap of ~18 meV with pronounced coherence peak is observed along the Brillouin zone diagonal direction. Remarkably, the finite superconducting gap persists across the entire Brillouin zone of the underlying Fermi surfaces, revealing the absence of gap nodes. An abrupt band renormalization, manifested as a kink in the energy-momentum dispersion at ~70 meV below the Fermi level, indicates an electron-boson coupling in the system. The simultaneous observation of a nodeless superconducting gap and electron-boson coupling provides crucial insights into the pairing symmetry and gluing mechanism in high-T$_c$ RP bilayer nickelates.

cond-mat.supr-con↗

arXiv2025

Mirror in the Model: Ad Banner Image Generation via Reflective Multi-LLM and Multi-modal Agents

Recent generative models such as GPT-4o have shown strong capabilities in producing high-quality images with accurate text rendering. However, commercial design tasks like advertising banners demand more than visual fidelity -- they require structured layouts, precise typography, consistent branding, and more. In this paper, we introduce MIMO (Mirror In-the-Model), an agentic refinement framework for automatic ad banner generation. MIMO combines a hierarchical multi-modal agent system (MIMO-Core) with a coordination loop (MIMO-Loop) that explores multiple stylistic directions and iteratively improves design quality. Requiring only a simple natural language based prompt and logo image as input, MIMO automatically detects and corrects multiple types of errors during generation. Experiments show that MIMO significantly outperforms existing diffusion and LLM-based baselines in real-world banner design scenarios.

cs.CV↗

arXiv2025

Superconducting gap and its Little-Parks like oscillations with high-order harmonics in lithium intercalated 1T-TiSe$_2$

The superconducting phase of doped 1T-TiSe$_2$ is a fruitful playground for exploring exotic quantum phenomena such as the anomalous metal state and spontaneously formed superconducting network. Here, we address these emergent states by studying the superconducting gap of lithium intercalated TiSe$_2$-a fundamental quantity that has remained unexplored so far. We fabricate a device that combines solid-state lateral lithium intercalation, resistance measurements and tunneling spectroscopy. We successfully probe the superconducting gap of TiSe$_2$ and reveal that the gap closing temperature well exceeds the transition temperature ($T_c$) expected from the Bardeen-Cooper-Schrieffer theory, indicating pronounced superconducting fluctuations even in a bulk-like system. Moreover, the symmetric gap persists even in the anomalous metal state, demonstrating the particle-hole symmetry of this exotic phase directly from the density of states. Finally, the superconducting gap shows magneto-oscillations with higher harmonics, attesting to a rather regular structure of the intrinsic superconducting network.

cond-mat.supr-con↗

arXiv2025

Kimi-VL Technical Report

We present Kimi-VL, an efficient open-source Mixture-of-Experts (MoE) vision-language model (VLM) that offers advanced multimodal reasoning, long-context understanding, and strong agent capabilities - all while activating only 2.8B parameters in its language decoder (Kimi-VL-A3B). Kimi-VL demonstrates strong performance across challenging domains: as a general-purpose VLM, Kimi-VL excels in multi-turn agent tasks (e.g., OSWorld), matching flagship models. Furthermore, it exhibits remarkable capabilities across diverse challenging vision language tasks, including college-level image and video comprehension, OCR, mathematical reasoning, and multi-image understanding. In comparative evaluations, it effectively competes with cutting-edge efficient VLMs such as GPT-4o-mini, Qwen2.5-VL-7B, and Gemma-3-12B-IT, while surpassing GPT-4o in several key domains. Kimi-VL also advances in processing long contexts and perceiving clearly. With a 128K extended context window, Kimi-VL can process diverse long inputs, achieving impressive scores of 64.5 on LongVideoBench and 35.1 on MMLongBench-Doc. Its native-resolution vision encoder, MoonViT, further allows it to see and understand ultra-high-resolution visual inputs, achieving 83.2 on InfoVQA and 34.5 on ScreenSpot-Pro, while maintaining lower computational cost for common tasks. Building upon Kimi-VL, we introduce an advanced long-thinking variant: Kimi-VL-Thinking-2506. Developed through long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL), the latest model exhibits strong long-horizon reasoning capabilities (64.0 on MMMU, 46.3 on MMMU-Pro, 56.9 on MathVision, 80.1 on MathVista, 65.2 on VideoMMMU) while obtaining robust general abilities. Code and models are publicly accessible at https://github.com/MoonshotAI/Kimi-VL.

cs.CV↗

arXiv2025

Theoretical Analysis of Near-Field MIMO Channel Capacity and Mid-Band Experimental Validation

With the increase of multiple-input-multiple-output (MIMO) array size and carrier frequency, near-field MIMO communications will become crucial in 6G wireless networks. Due to the increase of MIMO near-field range, the research of near-field MIMO capacity has aroused wide interest. In this paper, we focus on the theoretical analysis and empirical study of near-field MIMO capacity. First, the near-field channel model is characterized from the electromagnetic information perspective. Second, with the uniform planar array (UPA), the channel capacity based on effective degree of freedom (EDoF) is analyzed theoretically, and the closed-form analytical expressions are derived in detail. Finally, based on the numerical verification of near-field channel measurement experiment at 13 GHz band, we reveal that the channel capacity of UPA-type MIMO systems decreases continuously with the communication distance increasing. It can be observed that the near-field channel capacity gain is relatively obvious when large-scale MIMO is adopted at both receiving and transmitter ends, but the near-field channel capacity gain may be limited in the actual communication system with the small antenna array at receiving end. This work will give some reference to the near-field communication systems.

eess.SP↗

arXiv2025

Step-Audio-AQAA: a Fully End-to-End Expressive Large Audio Language Model

Large Audio-Language Models (LALMs) have significantly advanced intelligent human-computer interaction, yet their reliance on text-based outputs limits their ability to generate natural speech responses directly, hindering seamless audio interactions. To address this, we introduce Step-Audio-AQAA, a fully end-to-end LALM designed for Audio Query-Audio Answer (AQAA) tasks. The model integrates a dual-codebook audio tokenizer for linguistic and semantic feature extraction, a 130-billion-parameter backbone LLM and a neural vocoder for high-fidelity speech synthesis. Our post-training approach employs interleaved token-output of text and audio to enhance semantic coherence and combines Direct Preference Optimization (DPO) with model merge to improve performance. Evaluations on the StepEval-Audio-360 benchmark demonstrate that Step-Audio-AQAA excels especially in speech control, outperforming the state-of-art LALMs in key areas. This work contributes a promising solution for end-to-end LALMs and highlights the critical role of token-based vocoder in enhancing overall performance for AQAA tasks.

cs.SD↗

arXiv2025

Unveiling the Hidden: Movie Genre and User Bias in Spoiler Detection

Spoilers in movie reviews are important on platforms like IMDb and Rotten Tomatoes, offering benefits and drawbacks. They can guide some viewers' choices but also affect those who prefer no plot details in advance, making effective spoiler detection essential. Existing spoiler detection methods mainly analyze review text, often overlooking the impact of movie genres and user bias, limiting their effectiveness. To address this, we analyze movie review data, finding genre-specific variations in spoiler rates and identifying that certain users are more likely to post spoilers. Based on these findings, we introduce a new spoiler detection framework called GUSD (The code is available at https://github.com/AI-explorer-123/GUSD) (Genre-aware and User-specific Spoiler Detection), which incorporates genre-specific data and user behavior bias. User bias is calculated through dynamic graph modeling of review history. Additionally, the R2GFormer module combines RetGAT (Retentive Graph Attention Network) for graph information and GenreFormer for genre-specific aggregation. The GMoE (Genre-Aware Mixture of Experts) model further assigns reviews to specialized experts based on genre. Extensive testing on benchmark datasets shows that GUSD achieves state-of-the-art results. This approach advances spoiler detection by addressing genre and user-specific patterns, enhancing user experience on movie review platforms.

cs.IR↗

arXiv2025

Angle-resolved photoemission spectroscopy of superconducting (La,Pr)3Ni2O7/SrLaAlO4 heterostructures

Ruddlesden-Popper bilayer nickelate thin film superconductors discovered under ambient pressure enable vast possibilities for investigating electronic structures of the superconducting state. Here, we report angle-resolved photoemission spectroscopy (ARPES) measurements of 1, 2, and 3 unit-cell epitaxial La2.85Pr0.15Ni2O7 films grown on SrLaAlO4 substates, through pure-oxygen in situ sample transportation. Evidence obtained using photons with distinct probing depths shows that conduction is localized primarily at the first unit cell near the interface. Scanning transmission electron microscopy (STEM), together with energy-dispersive X-ray spectroscopy (EDS) and electron energy loss spectroscopy (EELS), indicates that interfacial Sr diffusion and pronounced p-d hybridization gradient may collectively account for the interfacial confinement of conduction. Fermi surface maps reveal hole doping compared to non-superconducting ambient-pressure bulk crystals. Measurements of dispersive band structures suggest the contributions from both Ni dx2-y2 and dz2 orbitals at the Fermi level. Density functional theory (DFT) + U calculations capture qualitative features of the ARPES results, consistent with a hole-doped scenario. These findings constrain theoretical models of the superconducting mechanism and suggest potential for enhancing superconductivity in nickelates under ambient pressure.

cond-mat.supr-con↗

arXiv2025

Multimodal Causal Reasoning Benchmark: Challenging Vision Large Language Models to Discern Causal Links Across Modalities

Multimodal Large Language Models (MLLMs) have showcased exceptional Chain-of-Thought (CoT) reasoning ability in complex textual inference tasks including causal reasoning. However, will these causalities remain straightforward when crucial hints hide in visual details? If not, what factors might influence cross-modal generalization? Whether we can effectively enhance their capacity for robust causal inference across both text and vision? Motivated by these, we introduce MuCR - a novel Multimodal Causal Reasoning benchmark that leverages synthetic siamese images and text pairs to challenge MLLMs. Additionally, we develop tailored metrics from multiple perspectives, including image-level match, phrase-level understanding, and sentence-level explanation, to comprehensively assess MLLMs' comprehension abilities. Our experiments reveal that current MLLMs fall short in multimodal causal reasoning compared to their performance in purely textual settings. Additionally, we find that identifying visual cues across images is key to effective cross-modal generalization. Finally, we propose a VcCoT strategy that better highlights visual cues, and our results confirm its efficacy in enhancing multimodal causal reasoning. The project is available at: https://github.com/Zhiyuan-Li-John/MuCR

cs.CV↗

arXiv2025

Highly Variable Quasar Candidates Selected from 4XMM-DR13 with Machine Learning

We present a sample of 12 quasar candidates with highly variable soft X-ray emission from the 4th XMM-newton Serendipitous Source Catalog (4XMM-DR13) using random forest. We obtained optical to mid-IR photometric data for the 4XMM-DR13 sources by correlating the sample with the SDSS DR18 photometric database and the AllWISE database. By cross-matching this sample with known spectral catalogs from the SDSS and LAMOST surveys, we obtained a training data set containing stars, galaxies, and quasars. The random forest algorithm was trained to classify the XMM-WISE-SDSS sample. We further filtered the classified quasar candidates with $\it{Gaia}$ proper motion to remove stellar contaminants. Finally, 53,992 quasar candidates have been classified, with 10,210 known quasars matched in SIMBAD. The quasar candidates have systematically lower X-ray fluxes than quasars in the training set, which indicates the classifier is helpful to single out fainter quasars. We constructed a sample of 12 sources from these quasars candidates which changed their soft X-ray fluxes by a factor of 10 over $\sim$ 20 years in the 4XMM-newton survey. Our selected highly variable quasar candidates extend the quasar sample, characterized by extreme soft X-ray variability, to the optically faint end with magnitudes around $r \sim 22$. None of the 12 sources were detected in ROSAT observations. Given the flux limit of ROSAT, the result suggests that quasars exhibiting variations of more than two orders of magnitudes are extremely rare.

astro-ph.GA↗

arXiv2025

Bolometric Superconducting Optical Nanoscopy (BOSON)

Superconducting transition-edge sensors are renowned for their extraordinary photon sensitivity and energy resolution, finding applications spanning quantum information, astronomy, and nanophotonics. Here, we report the development of BOlometric Superconducting Optical Nanoscopy (BOSON), a novel platform that integrates bolometric detection at the superconducting transition edges with near-field optical techniques. BOSON enables the mapping of photoinduced changes in superconductivity with unprecedented spatial resolution and photon sensitivity. By incorporating BOSON with low-dimensional materials, we achieved polariton imaging at nanowatt excitation levels--at least four orders of magnitude lower than the power typically required in prior near-field nanoscopy experiments. Our findings highlight the potential for BOSON to advance scanning probe based optical platforms to enable the detection of photons, polaritons, and Cooper pair dynamics at the nanoscale. This paves the way for quantum sensing applications using single-polariton detection and can offer deeper insights into quasiparticle dynamics.

quant-ph↗

arXiv2025

BUPTCMCC-6G-CMG+: A GBSM-Based ISAC Standard Channel Model Generator

Integrated sensing and communication (ISAC) has been recognized as the key technology in the vision of the sixth generation (6G) era. With the emergence of new concepts in mobile communications, the channel model is the prerequisite for system design and performance evaluation. Currently, 3GPP Release 19 is advancing the standardization of ISAC channel models. Nevertheless, a unified modeling framework has yet to be established. This paper provides a simulation diagram of ISAC channel modeling extended based on the Geometry-Based Stochastic Model (GBSM), compatible with existing 5G channel models and the latest progress in the 3rd Generation Partnership Project (3GPP) standardization. We first introduce the progress of the ISAC channel model standardization in general. Then, a concatenated channel modeling approach is presented considering the team's standardization proposals, which is implemented on the BUPTCMCC-6G-CMG+ channel model generator. We validated the model in cumulative probability density function (CDF) in statistical extension of angle and delay, and radar cross section (RCS). Simulation results show that the proposed model can realistically characterize the feature of channel concatenation and RCS within the ISAC channel.

eess.SP↗

arXiv2025

Photocurrent Nanoscopy of Quantum Hall Bulk

Understanding nanoscale electronic and thermal transport of two-dimensional (2D) electron systems in the quantum Hall regime, particularly in the bulk insulating state, poses considerable challenges. One of the primary difficulties arises from the presence of chiral edge channels, whose transport behavior obscures the investigation of the insulating bulk. Using near-field (NF) optical and photocurrent (PC) nanoscopy, we probe real-space variations of the optical and thermal dynamics of graphene in the quantum Hall regime without relying on complex sample or electrode geometries. Near the charge neutrality point (CNP), we detect strong optical and photothermal signals from resonant inter-Landau level (LL) magnetoexciton excitations between the 0th and +-1st LLs, which gradually weaken with increasing doping due to Pauli blocking. Interestingly, at higher doping levels and full integer LL fillings, photothermal signals reappear across the entire sample over a ~10-micrometer scale, indicating unexpectedly long cooling lengths and nonlocal photothermal heating through the insulating bulk. This observation suggests thermal conductivity persists for the localized states even as electronic transport is suppressed - a clear violation of the Wiedemann-Franz (WF) law. Our experiments provide novel insights into nanoscale thermal and electronic transport in incompressible 2D gases, highlighting the roles of magnetoexcitons and chiral edge states in the thermo-optoelectric dynamics of Dirac quantum Hall state.

cond-mat.mes-hall↗

arXiv2025

Open-Qwen2VL: Compute-Efficient Pre-Training of Fully-Open Multimodal LLMs on Academic Resources

The reproduction of state-of-the-art multimodal LLM pre-training faces barriers at every stage of the pipeline, including high-quality data filtering, multimodal data mixture strategies, sequence packing techniques, and training frameworks. We introduce Open-Qwen2VL, a fully open-source 2B-parameter Multimodal Large Language Model pre-trained efficiently on 29M image-text pairs using only 220 A100-40G GPU hours. Our approach employs low-to-high dynamic image resolution and multimodal sequence packing to significantly enhance pre-training efficiency. The training dataset was carefully curated using both MLLM-based filtering techniques (e.g., MLM-Filter) and conventional CLIP-based filtering methods, substantially improving data quality and training efficiency. The Open-Qwen2VL pre-training is conducted on academic level 8xA100-40G GPUs at UCSB on 5B packed multimodal tokens, which is 0.36% of 1.4T multimodal pre-training tokens of Qwen2-VL. The final instruction-tuned Open-Qwen2VL outperforms partially-open state-of-the-art MLLM Qwen2-VL-2B on various multimodal benchmarks of MMBench, SEEDBench, MMstar, and MathVista, indicating the remarkable training efficiency of Open-Qwen2VL. We open-source all aspects of our work, including compute-efficient and data-efficient training details, data filtering methods, sequence packing scripts, pre-training data in WebDataset format, FSDP-based training codebase, and both base and instruction-tuned model checkpoints. We redefine "fully open" for multimodal LLMs as the complete release of: 1) the training codebase, 2) detailed data filtering techniques, and 3) all pre-training and supervised fine-tuning data used to develop the model.

cs.CL↗