arXiv subjects

Kai Li

Publications and source records attributed to Kai Li.

At least 145 records · Page 8Linked to original sources

arXiv2025

A Fast and Lightweight Model for Causal Audio-Visual Speech Separation

Audio-visual speech separation (AVSS) aims to extract a target speech signal from a mixed signal by leveraging both auditory and visual (lip movement) cues. However, most existing AVSS methods exhibit complex architectures and rely on future context, operating offline, which renders them unsuitable for real-time applications. Inspired by the pipeline of RTFSNet, we propose a novel streaming AVSS model, named Swift-Net, which enhances the causal processing capabilities required for real-time applications. Swift-Net adopts a lightweight visual feature extraction module and an efficient fusion module for audio-visual integration. Additionally, Swift-Net employs Grouped SRUs to integrate historical information across different feature spaces, thereby improving the utilization efficiency of historical information. We further propose a causal transformation template to facilitate the conversion of non-causal AVSS models into causal counterparts. Experiments on three standard benchmark datasets (LRS2, LRS3, and VoxCeleb2) demonstrated that under causal conditions, our proposed Swift-Net exhibited outstanding performance, highlighting the potential of this method for processing speech in complex environments.

cs.SD↗

arXiv2025

Large Language Models in Operations Research: Methods, Applications, and Challenges

Operations research (OR) is a core methodology that supports complex system decision-making, with broad applications in transportation, supply chain management, and production scheduling. However, traditional approaches that rely on expert-driven modeling and manual parameter tuning often struggle with large-scale, dynamic, and multi-constraint problems, limiting scalability and real-time applicability. Large language models (LLMs), with capabilities in semantic understanding, structured generation, and reasoning control, offer new opportunities to overcome these challenges. They can translate natural language problem descriptions into mathematical models or executable code, generate heuristics, evolve algorithms, and directly solve optimization tasks. This shifts the paradigm from human-driven processes to intelligent human-AI collaboration. This paper systematically reviews progress in applying LLMs to OR, categorizing existing methods into three pathways: automatic modeling, auxiliary optimization, and direct solving. It also examines evaluation benchmarks and domain-specific applications, and highlights key challenges, including unstable semantic-to-structure mapping, fragmented research, limited generalization and interpretability, insufficient evaluation systems, and barriers to industrial deployment. Finally, it outlines potential research directions. Overall, LLMs demonstrate strong potential to reshape the OR paradigm by enhancing interpretability, adaptability, and scalability, paving the way for next-generation intelligent optimization systems.

cs.AI↗

arXiv2025

AudioMarathon: A Comprehensive Benchmark for Long-Context Audio Understanding and Efficiency in Audio LLMs

Processing long-form audio is a major challenge for Large Audio Language models (LALMs). These models struggle with the quadratic cost of attention ($O(N^2)$) and with modeling long-range temporal dependencies. Existing audio benchmarks are built mostly from short clips and do not evaluate models in realistic long context settings. To address this gap, we introduce AudioMarathon, a benchmark designed to evaluate both understanding and inference efficiency on long-form audio. AudioMarathon provides a diverse set of tasks built upon three pillars: long-context audio inputs with durations ranging from 90.0 to 300.0 seconds, which correspond to encoded sequences of 2,250 to 7,500 audio tokens, respectively, full domain coverage across speech, sound, and music, and complex reasoning that requires multi-hop inference. We evaluate state-of-the-art LALMs and observe clear performance drops as audio length grows. We also study acceleration techniques and analyze the trade-offs of token pruning and KV cache eviction. The results show large gaps across current LALMs and highlight the need for better temporal reasoning and memory-efficient architectures. We believe AudioMarathon will drive the audio and multimodal research community to develop more advanced audio understanding models capable of solving complex audio tasks.

cs.SD↗

arXiv2025

Photometric Analysis of 30 Contact Binaries in M31

M31, as the largest galaxy in the Local Group, is of significant importance for the study of stellar formation and evolution. Based on the data of 5,859 targets observed in M31 by Gu et al (2024), we selected 30 contact binaries by visual inspection for further study. Using the PHOEBE software and employing Bayesian optimization and Markov Chain Monte Carlo sampling, we determined the physical parameters of these 30 systems. The results show that 10 systems exhibit the O'Connell effect, which is well explained by introducing a dark spot on the primary star. 11 systems have mass ratios below 0.15, classifying them as extremely low mass ratio contact binaries, making them promising candidates for binary mergers. Six systems have primary star temperatures exceeding 10,000 K, classifying them as early-type contact binaries. The absolute physical parameters reveal that two contact binary systems contain massive stellar components, which may eventually evolve into compact binary star systems. To compare the effects of different galactic environments on the evolution of contact binaries, we constructed evolutionary diagrams for these 30 targets and for contact binaries in the Milky Way. The results show that, at the same mass, our targets have larger radii, higher luminosities, and lower orbital angular momenta than contact binaries in the Milky Way, indicating that they are at more advanced evolutionary stages. This may be attributed to the higher metallicity in M31 compared to the Milky Way.

astro-ph.SR↗

arXiv2025

Text2Move: Text-to-moving sound generation via trajectory prediction and temporal alignment

Human auditory perception is shaped by moving sound sources in 3D space, yet prior work in generative sound modelling has largely been restricted to mono signals or static spatial audio. In this work, we introduce a framework for generating moving sounds given text prompts in a controllable fashion. To enable training, we construct a synthetic dataset that records moving sounds in binaural format, their spatial trajectories, and text captions about the sound event and spatial motion. Using this dataset, we train a text-to-trajectory prediction model that outputs the three-dimensional trajectory of a moving sound source given text prompts. To generate spatial audio, we first fine-tune a pre-trained text-to-audio generative model to output temporally aligned mono sound with the trajectory. The spatial audio is then simulated using the predicted temporally-aligned trajectory. Experimental evaluation demonstrates reasonable spatial understanding of the text-to-trajectory model. This approach could be easily integrated into existing text-to-audio generative workflow and extended to moving sound generation in other spatial audio formats.

cs.SD↗

arXiv2025

Path-Controlled Secure Network Coding

Multicast for securely sharing confidential data among many users is becoming increasingly important. Currently, it relies on duplicate-and-forward routing and cryptographic methods based on computational security. However, these approaches neither attain multicast capacity of the network, nor ensure long-term security against advances in computing (information-theoretic security: ITS). Existing ITS solutions--quantum key distribution (QKD), physical layer security (PLS), and secure network coding (SNC)--still fail to enable scalable networks, as their underlying assumptions, such as trusted nodes and wiretap thresholds, gradually become invalid as the network grows. Here, we develop an efficient multi-tree multicast path-finding method to address this issue, integrating it with universal strongly ramp SNC. This system, path-controlled universal strongly ramp SNC (PUSNEC), can be overlaid onto QKD/PLS networks, enabling multicast capacity, ITS, and scalability. We derive the maximum leakage information to an eavesdropper under the probabilistic wiretap network assumption and demonstrate secure multicast in multi-hop networks through numerical simulations. Our quantitative analysis of the secrecyreliability tradeoff highlights a practical approach to achieving secure, reliable multicast on a global scale.

cs.IT↗

arXiv2025

A deep learning approach to inverse medium scattering: Learning regularizers from a direct imaging method

This paper aims to solve numerically the two-dimensional inverse medium scattering problem with far-field data. This is a challenging task due to the severe ill-posedness and strong nonlinearity of the inverse problem. As already known, it is necessary but also difficult numerically to employ an appropriate regularization strategy which effectively incorporates certain a priori information of the unknown scatterer to overcome the severe ill-posedness of the inverse problem. In this paper, we propose to use a deep learning approach to learn the a priori information of the support of the unknown scatterer from a direct imaging method. Based on the learned a priori information, we propose two inversion algorithms for solving the inverse problem. In the first one, the learned a priori information is incorporated into the projected Landweber method. In the second one, the learned a priori information is used to design the regularization functional for the regularized variational formulation of the inverse problem which is then solved with a traditional iteration algorithm. Extensive numerical experiments show that our inversion algorithms provide good reconstruction results even for the high contrast case and have a satisfactory generalization ability.

math.NA↗

arXiv2025

DragOSM: Extract Building Roofs and Footprints from Aerial Images by Aligning Historical Labels

Extracting polygonal roofs and footprints from remote sensing images is critical for large-scale urban analysis. Most existing methods rely on segmentation-based models that assume clear semantic boundaries of roofs, but these approaches struggle in off- nadir images, where the roof and footprint are significantly displaced, and facade pixels are fused with the roof boundary. With the increasing availability of open vector map annotations, e.g., OpenStreetMap, utilizing historical labels for off-nadir image annotation has become viable because remote sensing images are georeferenced once captured. However, these historical labels commonly suffer from significant positional discrepancies with new images and only have one annotation (roof or footprint), which fails to describe the correct structures of a building. To address these discrepancies, we first introduce a concept of an alignment token, which encodes the correction vector to guide the label correction. Based on this concept, we then propose Drag OpenStreetMap Labels (DragOSM), a novel model designed to align dislocated historical labels with roofs and footprints. Specifically, DragOSM formulates the label alignment as an interactive denoising process, modeling the positional discrepancy as a Gaussian distribution. During training, it learns to correct these errors by simulating misalignment with random Gaussian perturbations; during inference, it iteratively refines the positions of input labels. To validate our method, we further present a new dataset, Repairing Buildings in OSM (ReBO), comprising 179,265 buildings with both OpenStreetMap and manually corrected annotations across 5,473 images from 41 cities. Experimental results on ReBO demonstrate the effectiveness of DragOSM. Code, dataset, and trained models are publicly available at https://github.com/likaiucas/DragOSM.git.

cs.CV↗

arXiv2025

Observation of average topological phase in disordered Rydberg atom array

Topological phases have been extensively studied over the past two decades, primarily in quantum pure states, where they are protected by exact symmetries. Recently, numerous studies have theoretically demonstrated the existence of average symmetry-protected topological (SPT) phases in mixed quantum states, which naturally arise in real systems due to decoherence or disorder. Despite extensive experimental observations of exact SPT phases in various systems, ranging from solid-state materials to synthetic matters, average SPT phases are yet to be observed until this work. Here we report direct observations of disorder-induced many-body interacting average SPT phase in an atom array at half-filling, whereby random offsets to tweezer locations forming a lattice implement structural disorder, resulting in fluctuating long-range dipolar interactions between tweezer confined single atoms. The induced topological phase is vindicated by the spatially resolved atom-atom correlation functions for different forms of dimer compositions. The ground state degeneracy in disordered configurations is detected and compared to the regular lattice without disorder. By probing the quench dynamics of a highly excited state, we observe markedly slower decay of edge spin magnetization in comparison to the bulk spin, consistent with the presence of topologically protected edge modes in disordered lattices.

cond-mat.quant-gas↗

arXiv2025

A neural network model for quickly solving multiple-band light curves of contact binaries

The advent of large-scale photometric surveys has led to the discovery of over a million contact binary systems. Conventional light curve analysis methods are no longer adequate for handling such massive datasets. To address this challenge, we developed a neural network-based model capable of rapid analysis of multiple-band light curves of contact binaries. Our model can determine the fundamental physical parameters, including temperature and mass ratios, orbital inclination, potential, fillout factor, primary and secondary luminosities and radii, third light contribution, and spot parameters. Notably, unlike previous works, our model can simultaneously process multiple-band light curves and the four parameters of a starspot. The model's reliability was verified through analysis of the synthetic light curves generated by PHOEBE and the light curves of eight targets from \cite{2024ApJ...976..223W}'s work. The discrepancy distribution between the physical parameters determined by our model and true values for the synthetic light curves shows very good agreement. In addition, the physical parameters determined by our model and the corresponding light curve fits show remarkable consistency with \cite{2024ApJ...976..223W}'s results. By applying our model to OGLE contact binaries, physical parameters of 3,541 systems were obtained. We have packaged our model into an executable (CBLA.exe) file and archived it in the China-VO repository (https://doi.org/10.12149/101626). The software supports light curve analysis of 19 standard filters and allows for processing the data, whether from large-scale sky surveys or individual telescope observations.

astro-ph.SR↗

arXiv2025

SoloSpeech: Enhancing Intelligibility and Quality in Target Speech Extraction through a Cascaded Generative Pipeline

Target Speech Extraction (TSE) aims to isolate a target speaker's voice from a mixture of multiple speakers by leveraging speaker-specific cues, typically provided as auxiliary audio (a.k.a. cue audio). Although recent advancements in TSE have primarily employed discriminative models that offer high perceptual quality, these models often introduce unwanted artifacts, reduce naturalness, and are sensitive to discrepancies between training and testing environments. On the other hand, generative models for TSE lag in perceptual quality and intelligibility. To address these challenges, we present SoloSpeech, a novel cascaded generative pipeline that integrates compression, extraction, reconstruction, and correction processes. SoloSpeech features a speaker-embedding-free target extractor that utilizes conditional information from the cue audio's latent space, aligning it with the mixture audio's latent space to prevent mismatches. Evaluated on the widely-used Libri2Mix dataset, SoloSpeech achieves the new state-of-the-art intelligibility and quality in target speech extraction while demonstrating exceptional generalization on out-of-domain data and real-world scenarios.

eess.AS↗

arXiv2025

Mixture of Global and Local Experts with Diffusion Transformer for Controllable Face Generation

Controllable face generation poses critical challenges in generative modeling due to the intricate balance required between semantic controllability and photorealism. While existing approaches struggle with disentangling semantic controls from generation pipelines, we revisit the architectural potential of Diffusion Transformers (DiTs) through the lens of expert specialization. This paper introduces Face-MoGLE, a novel framework featuring: (1) Semantic-decoupled latent modeling through mask-conditioned space factorization, enabling precise attribute manipulation; (2) A mixture of global and local experts that captures holistic structure and region-level semantics for fine-grained controllability; (3) A dynamic gating network producing time-dependent coefficients that evolve with diffusion steps and spatial locations. Face-MoGLE provides a powerful and flexible solution for high-quality, controllable face generation, with strong potential in generative modeling and security applications. Extensive experiments demonstrate its effectiveness in multimodal and monomodal face generation settings and its robust zero-shot generalization capability. Project page is available at https://github.com/XavierJiezou/Face-MoGLE.

cs.CV↗

arXiv2025

IRSAMap:Towards Large-Scale, High-Resolution Land Cover Map Vectorization

With the enhancement of remote sensing image resolution and the rapid advancement of deep learning, land cover mapping is transitioning from pixel-level segmentation to object-based vector modeling. This shift demands more from deep learning models, requiring precise object boundaries and topological consistency. However, existing datasets face three main challenges: limited class annotations, small data scale, and lack of spatial structural information. To overcome these issues, we introduce IRSAMap, the first global remote sensing dataset for large-scale, high-resolution, multi-feature land cover vector mapping. IRSAMap offers four key advantages: 1) a comprehensive vector annotation system with over 1.8 million instances of 10 typical objects (e.g., buildings, roads, rivers), ensuring semantic and spatial accuracy; 2) an intelligent annotation workflow combining manual and AI-based methods to improve efficiency and consistency; 3) global coverage across 79 regions in six continents, totaling over 1,000 km; and 4) multi-task adaptability for tasks like pixel-level classification, building outline extraction, road centerline extraction, and panoramic segmentation. IRSAMap provides a standardized benchmark for the shift from pixel-based to object-based approaches, advancing geographic feature automation and collaborative modeling. It is valuable for global geographic information updates and digital twin construction. The dataset is publicly available at https://github.com/ucas-dlg/IRSAMap

cs.CV↗

arXiv2025

BSN-II: The First Light Curve Study of Eight Total Eclipsing Contact Binary Stars with Shallow Fillout Factors

This study provides the first comprehensive analysis of eight total-eclipse W Ursae Majoris-type contact binary systems. Ground-based photometric multiband observations were conducted at a Mexican observatory, and new times of minima were extracted. The O-C analysis reveals that four of our target binaries exhibit a long-term increase in their orbital periods, while the others show a long-term decrease in their orbital periods. We analyzed the light curves using the PHOEBE Python code and BSN application. Among the target systems, two required the inclusion of a cold starspot on one of the components to achieve an adequate fit. The light curve analysis revealed that the target systems exhibit a shallow fillout factor. Absolute parameters were estimated using the Gaia DR3 parallax and astrophysics equations. Considering the effective temperatures and component masses, each system was classified as either the A- or W-subtype. The stellar evolution of the systems was represented through the mass-radius and mass-luminosity diagrams. Additionally, we calculated the initial masses of the companion stars and the total mass lost for each target system.

astro-ph.SR↗

arXiv2025

Ethereum Crypto Wallets under Address Poisoning: How Usable and Secure Are They?

Blockchain address poisoning is an emerging phishing attack that crafts "similar-looking" transfer records in the victim's transaction history, which aims to deceive victims and lure them into mistakenly transferring funds to the attacker. Recent works have shown that millions of Ethereum users were targeted and lost over 100 million US dollars. Ethereum crypto wallets, serving users in browsing transaction history and initiating transactions to transfer funds, play a central role in deploying countermeasures to mitigate the address poisoning attack. However, whether they have done so remains an open question. To fill the research void, in this paper, we design experiments to simulate address poisoning attacks and systematically evaluate the usability and security of 53 popular Ethereum crypto wallets. Our evaluation shows that there exist communication failures between 12 wallets and their transaction activity provider, which renders them unable to download the users' transaction history. Besides, our evaluation also shows that 16 wallets pose a high risk to their users due to displaying fake token phishing transfers. Moreover, our further analysis suggests that most wallets rely on transaction activity providers to filter out phishing transfers. However, their phishing detection capability varies. Finally, we found that only three wallets throw an explicit warning message when users attempt to transfer to the phishing address, implying a significant gap within the broader Ethereum crypto wallet community in protecting users from address poisoning attacks. Overall, our work shows that more efforts are needed by the Ethereum crypto wallet developer community to achieve the highest usability and security standard. Our bug reports have been acknowledged by the developer community, who are currently developing mitigation solutions.

cs.CR↗

arXiv2025

GDSR: Global-Detail Integration through Dual-Branch Network with Wavelet Losses for Remote Sensing Image Super-Resolution

In recent years, deep neural networks, including Convolutional Neural Networks, Transformers, and State Space Models, have achieved significant progress in Remote Sensing Image (RSI) Super-Resolution (SR). However, existing SR methods typically overlook the complementary relationship between global and local dependencies. These methods either focus on capturing local information or prioritize global information, which results in models that are unable to effectively capture both global and local features simultaneously. Moreover, their computational cost becomes prohibitive when applied to large-scale RSIs. To address these challenges, we introduce the novel application of Receptance Weighted Key Value (RWKV) to RSI-SR, which captures long-range dependencies with linear complexity. To simultaneously model global and local features, we propose the Global-Detail dual-branch structure, GDSR, which performs SR by paralleling RWKV and convolutional operations to handle large-scale RSIs. Furthermore, we introduce the Global-Detail Reconstruction Module (GDRM) as an intermediary between the two branches to bridge their complementary roles. In addition, we propose the Dual-Group Multi-Scale Wavelet Loss, a wavelet-domain constraint mechanism via dual-group subband strategy and cross-resolution frequency alignment for enhanced reconstruction fidelity in RSI-SR. Extensive experiments under two degradation methods on several benchmarks, including AID, UCMerced, and RSSRD-QH, demonstrate that GSDR outperforms the state-of-the-art Transformer-based method HAT by an average of 0.09 dB in PSNR, while using only 63% of its parameters and 51% of its FLOPs, achieving an inference speed 3.2 times faster.

eess.IV↗

arXiv2025

Advances in Speech Separation: Techniques, Challenges, and Future Trends

The field of speech separation, addressing the "cocktail party problem", has seen revolutionary advances with DNNs. Speech separation enhances clarity in complex acoustic environments and serves as crucial pre-processing for speech recognition and speaker recognition. However, current literature focuses narrowly on specific architectures or isolated approaches, creating fragmented understanding. This survey addresses this gap by providing systematic examination of DNN-based speech separation techniques. Our work differentiates itself through: (I) Comprehensive perspective: We systematically investigate learning paradigms, separation scenarios with known/unknown speakers, comparative analysis of supervised/self-supervised/unsupervised frameworks, and architectural components from encoders to estimation strategies. (II) Timeliness: Coverage of cutting-edge developments ensures access to current innovations and benchmarks. (III) Unique insights: Beyond summarization, we evaluate technological trajectories, identify emerging patterns, and highlight promising directions including domain-robust frameworks, efficient architectures, multimodal integration, and novel self-supervised paradigms. (IV) Fair evaluation: We provide quantitative evaluations on standard datasets, revealing true capabilities and limitations of different methods. This comprehensive survey serves as an accessible reference for experienced researchers and newcomers navigating speech separation's complex landscape.

cs.SD↗

arXiv2025

Repurposing 2D Diffusion Models with Gaussian Atlas for 3D Generation

Recent advances in text-to-image diffusion models have been driven by the increasing availability of paired 2D data. However, the development of 3D diffusion models has been hindered by the scarcity of high-quality 3D data, resulting in less competitive performance compared to their 2D counterparts. To address this challenge, we propose repurposing pre-trained 2D diffusion models for 3D object generation. We introduce Gaussian Atlas, a novel representation that utilizes dense 2D grids, enabling the fine-tuning of 2D diffusion models to generate 3D Gaussians. Our approach demonstrates successful transfer learning from a pre-trained 2D diffusion model to a 2D manifold flattened from 3D structures. To support model training, we compile GaussianVerse, a large-scale dataset comprising 205K high-quality 3D Gaussian fittings of various 3D objects. Our experimental results show that text-to-image diffusion models can be effectively adapted for 3D content generation, bridging the gap between 2D and 3D modeling.

cs.CV↗