arXiv subjects

Hao Zhang

Publications and source records attributed to Hao Zhang.

At least 379 records · Page 21Linked to original sources

arXiv2025

AgentArcEval: An Architecture Evaluation Method for Foundation Model based Agents

The emergence of foundation models (FMs) has enabled the development of highly capable and autonomous agents, unlocking new application opportunities across a wide range of domains. Evaluating the architecture of agents is particularly important as the architectural decisions significantly impact the quality attributes of agents given their unique characteristics, including compound architecture, autonomous and non-deterministic behaviour, and continuous evolution. However, these traditional methods fall short in addressing the evaluation needs of agent architecture due to the unique characteristics of these agents. Therefore, in this paper, we present AgentArcEval, a novel agent architecture evaluation method designed specially to address the complexities of FM-based agent architecture and its evaluation. Moreover, we present a catalogue of agent-specific general scenarios, which serves as a guide for generating concrete scenarios to design and evaluate the agent architecture. We demonstrate the usefulness of AgentArcEval and the catalogue through a case study on the architecture evaluation of a real-world tax copilot, named Luna.

cs.SE↗

arXiv2025

Fully automatic fabrication of fibre Bragg gratings using an AI-powered femtosecond laser inscription system

Fibre Bragg gratings (FBGs) are widely used in optical sensing and communication systems. Femtosecond laser inscription (FLI) enables hydrogen-free, thermally stable, high-resolution, and complex structures of FBG fabrication, but its practical application is limited by manual operation, low throughput, and sensitivity to laser alignment. In this study, we present an AI-powered FLI system that enables automated, stable, and efficient FBG fabrication. By integrating a Multi-Layer Perceptron (MLP) model for real-time fabrication position correction, the system maintains precise laser alignment (-0.6 to 0.2 microns of the fibre core plane) and ensures consistent processing. Strong and weak FBGs were fabricated in different types of fibres, and their spectral characteristics-including central wavelength, reflectivity, and FWHM-exhibited high stability and repeatability. The results demonstrate that the proposed AI-powered FLI system significantly reduces manual intervention while achieving reliable FBG performance. This approach holds great promise for scalable, high-throughput FBG production and can be extended to the fabrication of arbitrary FBG structures across various fibre types. With further training and model refinement, the AI-powered FLI provides a scalable and intelligent platform for next-generation automated FBG manufacturing.

physics.optics↗

arXiv2025

Efficient Long-context Language Model Training by Core Attention Disaggregation

We present core attention disaggregation (CAD), a technique that improves long-context large language model training by decoupling the core attention computation, softmax(QK^T)V, from the rest of the model and executing it on a separate pool of devices. In existing systems, core attention is colocated with other layers; at long context lengths, its quadratic compute growth compared to the near-linear growth of other components causes load imbalance and stragglers across data and pipeline parallel groups. CAD is enabled by two observations. First, core attention is stateless: it has no trainable parameters and only minimal transient data, so balancing reduces to scheduling compute-bound tasks. Second, it is composable: modern attention kernels retain high efficiency when processing fused batches of token-level shards with arbitrary lengths. CAD partitions core attention into token-level tasks and dispatches them to dedicated attention servers, which dynamically rebatch tasks to equalize compute without sacrificing kernel efficiency. We implement CAD in a system called DistCA, which uses a ping-pong execution scheme to fully overlap communication with computation and in-place execution on attention servers to reduce memory use. On 512 H200 GPUs and context lengths up to 512k tokens, DistCA improves end-to-end training throughput by up to 1.35x, eliminates data and pipeline parallel stragglers, and achieves near-perfect compute and memory balance.

cs.LG↗

arXiv2025

Total instanton restriction via multiverse interference: Noncompact gauge theories and (-1)-form symmetries

In this note we consider examples of decomposition (in which a local QFT is equivalent to a disjoint union of multiple independent theories, known as universes) where there is a continuous familiy of universes, rather than a finite or countably infinite collection. In particular, this allows us to consistently eliminate all instantons in a local QFT via a suitable topological gauging of the (-1)-form symmetry. In two-dimensional U(1) gauge theories, this is equivalent to changing the gauge group to R. This makes both locality as well as the instanton restriction explicit. We apply this to clarify the Gross-Taylor string interpretation of the decomposition of two-dimensional pure Yang-Mills. We also apply decomposition to study two-dimensional R gauge theories, such as the pure R Maxwell theory, and two-dimensional supersymmetric gauged linear sigma models whose gauge groups have factors of R. In that context, we find that analogues of the Witten effect for dyons, here rotating between universes, play a role in relating anomalies of the individual universes to (different) anomalies in the disjoint union. Finally, we discuss limits of the Tanizaki-Unsal construction, which accomplish instanton restriction by topologically gauging a Q/Z (-1)-form symmetry, and speculate in two-dimensional theories on possible interpretations of those limits in terms of the adelic solenoid.

hep-th↗

arXiv2025

Can Representation Gaps Be the Key to Enhancing Robustness in Graph-Text Alignment?

Representation learning on text-attributed graphs (TAGs) integrates structural connectivity with rich textual semantics, enabling applications in diverse domains. Current methods largely rely on contrastive learning to maximize cross-modal similarity, assuming tighter coupling between graph and text representations improves transfer performance. However, our empirical analysis reveals that both natural gap expansion and forced gap reduction result in performance degradation by disrupting pre-trained knowledge structures and impairing generalization. This arises from the geometric incompatibility between encoders, where graph encoders capture topological patterns, while text encoders capture semantic structures. Over-alignment compresses these distinct spaces into shared subspaces, causing structure collapse that diminishes both topological reasoning and semantic understanding. We propose \textbf{LLM4GTA}, a gap-aware alignment framework that preserves representation gaps as geometric necessities for maintaining modality-specific knowledge and improving transfer performance. LLM4GTA includes an adaptive gap preservation module to prevent over-alignment by monitoring similarity evolution and an intra-modal compensation mechanism that boosts discriminative power using auxiliary classifiers in graph space. Extensive experiments show significant improvements over existing methods in zero-shot and few-shot scenarios.

cs.GR↗

arXiv2025

Spatial two-grid compact difference scheme for two-dimensional nonlinear diffusion-wave equations with variable exponent

This paper presents a spatial two-grid (STG) compact difference scheme for a two-dimensional (2D) nonlinear diffusion-wave equation with variable exponent, which describes, e.g., the propagation of mechanical diffusive waves in viscoelastic media with varying material properties. Following the idea of the convolution approach, the diffusion-wave model is first transformed into an equivalent formulation. A fully discrete scheme is then developed by applying a compact difference approximation in space and combining the averaged product integration rule with linear interpolation quadrature in time. An efficient high-order two-grid algorithm is constructed by solving a small-scale nonlinear system on the coarse grid and a large-scale linearized system on the fine grid, where the bicubic spline interpolation operator is used to project coarse-grid solutions to the fine grid. Under mild assumptions on the variable exponent $α(t)$, the stability and convergence of the STG compact difference scheme are rigorously established. Numerical experiments are finally presented to verify the accuracy and efficiency of the proposed method.

math.NA↗

arXiv2025

SANA-Video: Efficient Video Generation with Block Linear Diffusion Transformer

We introduce SANA-Video, a small diffusion model that can efficiently generate videos up to 720x1280 resolution and minute-length duration. SANA-Video synthesizes high-resolution, high-quality and long videos with strong text-video alignment at a remarkably fast speed, deployable on RTX 5090 GPU. Two core designs ensure our efficient, effective and long video generation: (1) Linear DiT: We leverage linear attention as the core operation, which is more efficient than vanilla attention given the large number of tokens processed in video generation. (2) Constant-Memory KV cache for Block Linear Attention: we design block-wise autoregressive approach for long video generation by employing a constant-memory state, derived from the cumulative properties of linear attention. This KV cache provides the Linear DiT with global context at a fixed memory cost, eliminating the need for a traditional KV cache and enabling efficient, minute-long video generation. In addition, we explore effective data filters and model training strategies, narrowing the training cost to 12 days on 64 H100 GPUs, which is only 1% of the cost of MovieGen. Given its low cost, SANA-Video achieves competitive performance compared to modern state-of-the-art small diffusion models (e.g., Wan 2.1-1.3B and SkyReel-V2-1.3B) while being 16x faster in measured latency. Moreover, SANA-Video can be deployed on RTX 5090 GPUs with NVFP4 precision, accelerating the inference speed of generating a 5-second 720p video from 71s to 29s (2.4x speedup). In summary, SANA-Video enables low-cost, high-quality video generation.

cs.CV↗

arXiv2025

Scaling Language-Centric Omnimodal Representation Learning

Recent multimodal embedding approaches leveraging multimodal large language models (MLLMs) fine-tuned with contrastive learning (CL) have shown promising results, yet the underlying reasons behind their superiority remain underexplored. This work argues that a crucial advantage of MLLM-based approaches stems from implicit cross-modal alignment achieved during generative pretraining, where the language decoder learns to exploit multimodal signals within a shared representation space for generating unimodal outputs. Through analysis of anisotropy and kernel similarity structure, we empirically confirm that latent alignment emerges within MLLM representations, allowing CL to serve as a lightweight refinement stage. Leveraging this insight, we propose a Language-Centric Omnimodal Embedding framework, termed LCO-Emb. Extensive experiments across diverse backbones and benchmarks demonstrate its effectiveness, achieving state-of-the-art performance across modalities. Furthermore, we identify a Generation-Representation Scaling Law (GRSL), showing that the representational capabilities gained through contrastive refinement scales positively with the MLLM's generative capabilities. This suggests that improving generative abilities evolves as an effective paradigm for enhancing representation quality. We provide a theoretical explanation of GRSL, which formally links the MLLM's generative quality to the upper bound on its representation performance, and validate it on a challenging, low-resource visual-document retrieval task, showing that continual generative pretraining before CL can further enhance the potential of a model's embedding capabilities. Codes, models, and resources are available at https://github.com/LCO-Embedding/LCO-Embedding.

cs.CL↗

arXiv2025

MultiCOIN: Multi-Modal COntrollable Video INbetweening

Video inbetweening creates smooth and natural transitions between two image frames, making it an indispensable tool for video editing and long-form video synthesis. Existing works in this domain are unable to generate large, complex, or intricate motions. In particular, they cannot accommodate the versatility of user intents and generally lack fine control over the details of intermediate frames, leading to misalignment with the creative mind. To fill these gaps, we introduce MultiCOIN, a video inbetweening framework that allows multi-modal controls, including depth transition and layering, motion trajectories, text prompts, and target regions for movement localization, while achieving a balance between flexibility, ease of use, and precision for fine-grained video interpolation. To achieve this, we adopt the Diffusion Transformer (DiT) architecture as our video generative model, due to its proven capability to generate high-quality long videos. To ensure compatibility between DiT and our multi-modal controls, we map all motion controls into a common sparse and user-friendly point-based representation as the video/noise input. Further, to respect the variety of controls which operate at varying levels of granularity and influence, we separate content controls and motion controls into two branches to encode the required features before guiding the denoising process, resulting in two generators, one for motion and the other for content. Finally, we propose a stage-wise training strategy to ensure that our model learns the multi-modal controls smoothly. Extensive qualitative and quantitative experiments demonstrate that multi-modal controls enable a more dynamic, customizable, and contextually accurate visual narrative.

cs.CV↗

arXiv2025

Towards Quantum Simulations of Sphaleron Dynamics at Colliders

Sphaleron dynamics in the Standard Model at high-energy particle collisions remains experimentally unobserved, with theoretical predictions hindered by its nonperturbative real-time nature. In this work, we investigate a quantum simulation approach to this challenge. Taking the $1+1$D $O(3)$ model as a protocol towards studying dynamics of sphaleron in the electroweak theory, we identify the sphaleron configuration and establish lattice parameters that reproduce continuum sphaleron energies with controlled precision. We then develop quantum algorithms to simulate sphaleron evolutions where quantum effects can be included. This work lays the ground to establish quantum simulations for studying the interaction between classical topological objects and particles in the quantum field theory that are usually inaccessible to classical methods and computations.

hep-ph↗

arXiv2025

ReasonMed: A 370K Multi-Agent Generated Dataset for Advancing Medical Reasoning

Reasoning-based large language models have excelled in mathematics and programming, yet their potential in knowledge-intensive medical question answering remains underexplored and insufficiently validated in clinical contexts. To bridge this gap, we introduce ReasonMed, the largest medical reasoning dataset to date, comprising 370k high-quality examples distilled from 1.75 million initial reasoning paths generated by complementary LLMs and curated through a cost-efficient easy-medium-difficult (EMD) pipeline. ReasonMed is built through a multi-agent generation, verification, and refinement process, in which an Error Refiner improves reasoning paths by correcting error-prone steps identified by a verifier. Using ReasonMed, we investigate effective strategies for training medical reasoning models and find that integrating detailed CoT reasoning with concise answer summaries yields the most robust fine-tuning results. Models trained on ReasonMed set a new benchmark: ReasonMed-7B surpasses the prior best sub-10B models by 4.17% and even exceeds LLaMA3.1-70B on PubMedQA by 4.60%. When scaled to ReasonMed-14B, it remains highly competitive, underscoring consistent scaling potential. The codes and datasets are available at https://github.com/YuSun-Work/ReasonMed.

cs.CL↗

arXiv2025

ElasWave: An Elastic-Native System for Scalable Hybrid-Parallel Training

Large-scale LLM pretraining now runs across $10^5$--$10^6$ accelerators, making failures routine and elasticity mandatory. We posit that an elastic-native training system must jointly deliver (i) parameter consistency, (ii) low mean time to recovery (MTTR), (iii) high post-change throughput, and (iv) computation consistency. No prior system achieves all four simultaneously. To achieve these goals, we present ElasWave, which delivers per-step fault tolerance via multi-dimensional scheduling across graph, dataflow, DVFS, and RNG. ElasWave reshapes and reshards micro-batches while preserving the global batch size and gradient scale. It performs online pipeline resharding with asynchronous parameter migration and interleaves ZeRO partitions, reducing parameter recovery processes to disjoint rank-to-rank transfers. It further leverages DVFS to absorb pipeline bubbles and reshards RNG to keep computation consistency. Together, a dynamic communicator enables in-place communication group edits, while per-step in-memory snapshots support online verification and redistribution. We evaluate ElasWave on 96 NPUs and benchmark it against state-of-the-art baselines: throughput improves by $1.35\times$ over ReCycle and $1.60\times$ over TorchFT; communicator recovery completes within one second (up to $82\times/3.6\times$ faster than full/partial rebuilds); migration MTTR drops by as much as $51\%$; and convergence deviation is reduced by approximately $78\%$.

cs.DC↗

arXiv2025

Maritime Communication in Evaporation Duct Environment with Ship Trajectory Optimization

In maritime wireless networks, the evaporation duct effect has been known as a preferable condition for long-range transmissions. However, how to effectively utilize the duct effect for efficient communication design is still open for investigation. In this paper, we consider a typical scenario of ship-to-shore data transmission, where a ship collects data from multiple oceanographic buoys, sails from one to another, and transmits the collected data back to a terrestrial base station during its voyage. A novel framework, which exploits priori information of the channel gain map in the presence of evaporation duct, is proposed to minimize the data transmission time and the sailing time by optimizing the ship's trajectory. To this end, a multi-objective optimization problem is formulated and is further solved by a dynamic population PSO-integrated NSGA-II algorithm. Through simulations, it is demonstrated that, compared to the benchmark scheme which ignores useful information of the evaporation duct, the proposed scheme can effectively reduce both the data transmission time and the sailing time.

eess.SP↗

arXiv2025

Implementation of multiparticle quantum speed limits on observables

The energy-time uncertainty relation limits the maximum speed of quantum system evolution and is crucial for determining whether quantum tasks can be accelerated. However, multiparticle quantum speed limits have not been experimentally explored. In this work, we experimentally verify that both multiparticles and entanglement can accelerate the quantum speed on observables in two-particle systems based on ultrahigh precision control of quantum evolution time. Furthermore, we experimentally prove that the initial quantum state plays a critical role in the quantum speed limits of the entangled systems. In addition, we experimentally demonstrate that the upper bound and lower bound of the quantum speed are workable even in a nonunitary Markovian open system with two photons. The results obtained based on two-photon experiments have been shown to be generalizable to more particles. Our work facilitates the characterization of the dynamic transient properties of complex quantum systems and the control of the quantum speed of large-scale quantum systems.

quant-ph↗

arXiv2025

TrimTokenator: Towards Adaptive Visual Token Pruning for Large Multimodal Models

Large Multimodal Models (LMMs) have achieved significant success across various tasks. These models usually encode visual inputs into dense token sequences, which are then concatenated with textual tokens and jointly processed by a language model. However, the increased token count substantially raises computational and memory costs during inference. Token pruning has emerged as a promising approach to address this issue. Existing token pruning methods often rely on costly calibration or suboptimal importance metrics, leading to redundant retained tokens. In this paper, we analyze the redundancy differences between visual and textual tokens and propose pruning exclusively on visual tokens. Based on this, we propose a visual token pruning strategy that explicitly preserves both cross-modal alignment and intra-modal informational diversity. We introduce a mutual information-based token pruning strategy that removes visual tokens semantically misaligned with textual tokens, effectively preserving the alignment between the visual and textual modalities. To further improve the representational quality of the retained tokens, we additionally prune redundant visual tokens by maximizing the expected pairwise distances in the embedding space, which is solved efficiently with a greedy algorithm. Extensive experiments demonstrate that our method maintains strong performance while reducing tokens by 88.9% on models such as LLaVA-1.5-7B and LLaVA-NEXT-7B, resulting in a 56.7% improvement in inference speed.

cs.CV↗

arXiv2025

Towards Interpretable and Inference-Optimal COT Reasoning with Sparse Autoencoder-Guided Generation

We propose a novel method that leverages sparse autoencoders (SAEs) and clustering techniques to analyze the internal token representations of large language models (LLMs) and guide generations in mathematical reasoning tasks. Our approach first trains an SAE to generate sparse vector representations for training tokens, then applies k-means clustering to construct a graph where vertices represent token clusters and weighted edges capture sequential token transitions. Using this graph, we define an edge-weight based reward function to quantify adherence to established reasoning traces, thereby identifying exploitative reasoning trajectories. Additionally, we measure generation diversity from clustering to assess the extent of exploration. Our findings indicate that balancing both exploitation and exploration is crucial for achieving high accuracy in mathematical reasoning tasks. During generation, the SAE can serve as a scalable reward model to guide generations, ensuring a balanced trade-off between exploitation and exploration. This prevents extreme behaviors in either direction, ultimately fostering a higher-quality reasoning process in LLMs.

cs.AI↗

arXiv2025

Multimode Phonon-Polaritons in Lead-Halide Perovskites in the Ultrastrong Coupling Regime

Phonons play a central role in fundamental solid-state phenomena, including superconductivity, Raman scattering, and symmetry-breaking phases. Harnessing phonons to control these effects and enable quantum technologies is therefore of great interest. However, most existing phonon control strategies rely on external driving fields or anharmonic interactions, limiting their applicability. Here, we realize multimode ultrastrong light--matter coupling and theoretically show the modulation of phonon emission. This regime is realized by coupling two optical phonon modes in lead halide perovskites to a nanoslot array functioning as a single-mode cavity. The small mode volume of the nanoslots enables high coupling strengths in the phonon-polariton system. We show theoretically that the nanoslot resonator mediates an effective interaction between phonon modes, leading to superthermal phonon bunching in thermal equilibrium between distinct modes. Our findings are well described by a multimode Hopfield model. This work establishes a pathway for engineering phononic properties for light-harvesting and light-emitting technologies.

quant-ph↗

arXiv2025

Strongly Entangled Kondo and Kagome Lattices and the Emergent Magnetic Ground State in Heavy-Fermion Kagome Metal YbV$_6$Sn$_6$

Applying angle-resolved photoemission spectroscopy and density functional theory calculations, we present compelling spectroscopic evidence demonstrating the intertwining and mutual interaction between the Kondo and kagome sublattices in heavy-fermion intermetallic compound YbV$_6$Sn$_6$. We reveal the Yb 4$f$-derived states near the Fermi level, along with the presence of bulk kagome bands and topological surface states. We unveil strong interactions between the 4$f$ and itinerant electrons, where the kagome bands hosting the Dirac fermions and van Hove singularities predominate. Such findings are well described using a $c$-$f$ hybridization model. On the other hand, our systematic characterization of magnetic properties demonstrates an unusually enhanced antiferromagnetic ordering, where the kagome-derived van Hove singularities near $E_F$ play a vital role in determining the unconventional nature of the Ruderman-Kittel-Kasuya-Yosida interaction and Kondo coupling. These unique kagome-state-mediated exchange interactions have never been reported before and could lead to a novel phase diagram and various quantum critical behaviors in YbV$_6$Sn$_6$ and its siblings. Our results not only expand the family of exotic quantum phases entangled with kagome structure to the strongly correlated regime, but also establish YbV$_6$Sn$_6$ as an unprecedented platform to explore unconventional many-body physics beyond the standard Kondo picture.

cond-mat.str-el↗