arXiv subjects

Hao Wang

Publications and source records attributed to Hao Wang.

At least 595 records · Page 33Linked to original sources

arXiv2025

Unified Open-World Segmentation with Multi-Modal Prompts

In this work, we present COSINE, a unified open-world segmentation model that consolidates open-vocabulary segmentation and in-context segmentation with multi-modal prompts (e.g., text and image). COSINE exploits foundation models to extract representations for an input image and corresponding multi-modal prompts, and a SegDecoder to align these representations, model their interaction, and obtain masks specified by input prompts across different granularities. In this way, COSINE overcomes architectural discrepancies, divergent learning objectives, and distinct representation learning strategies of previous pipelines for open-vocabulary segmentation and in-context segmentation. Comprehensive experiments demonstrate that COSINE has significant performance improvements in both open-vocabulary and in-context segmentation tasks. Our exploratory analyses highlight that the synergistic collaboration between using visual and textual prompts leads to significantly improved generalization over single-modality approaches.

cs.CV↗

arXiv2025

High-Fidelity Simulated Data Generation for Real-World Zero-Shot Robotic Manipulation Learning with Gaussian Splatting

The scalability of robotic learning is fundamentally bottlenecked by the significant cost and labor of real-world data collection. While simulated data offers a scalable alternative, it often fails to generalize to the real world due to significant gaps in visual appearance, physical properties, and object interactions. To address this, we propose RoboSimGS, a novel Real2Sim2Real framework that converts multi-view real-world images into scalable, high-fidelity, and physically interactive simulation environments for robotic manipulation. Our approach reconstructs scenes using a hybrid representation: 3D Gaussian Splatting (3DGS) captures the photorealistic appearance of the environment, while mesh primitives for interactive objects ensure accurate physics simulation. Crucially, we pioneer the use of a Multi-modal Large Language Model (MLLM) to automate the creation of physically plausible, articulated assets. The MLLM analyzes visual data to infer not only physical properties (e.g., density, stiffness) but also complex kinematic structures (e.g., hinges, sliding rails) of objects. We demonstrate that policies trained entirely on data generated by RoboSimGS achieve successful zero-shot sim-to-real transfer across a diverse set of real-world manipulation tasks. Furthermore, data from RoboSimGS significantly enhances the performance and generalization capabilities of SOTA methods. Our results validate RoboSimGS as a powerful and scalable solution for bridging the sim-to-real gap.

cs.RO↗

arXiv2025

Lost in the Middle: An Emergent Property from Information Retrieval Demands in LLMs

The performance of Large Language Models (LLMs) often degrades when crucial information is in the middle of a long context, a "lost-in-the-middle" phenomenon that mirrors the primacy and recency effects in human memory. We propose that this behavior is not simply a flaw indicative of information loss but an adaptation to different information retrieval demands during pre-training: some tasks require uniform recall across the entire input (a long-term memory demand), while others prioritize the most recent information (a short-term memory demand). Consistent with this view, we show that this U-shaped performance curve emerges when LLMs (GPT-2 and Llama variants) are trained from scratch on two simple human memory paradigms simulating long-term and short-term memory demands. Our analysis reveals that while the recency effect directly aligns with short-term memory demand in the training data, the primacy effect is induced by the uniform long-term memory demand and is additionally influenced by the model's autoregressive properties and the formation of attention sinks. Our main findings from simple human memory paradigms also generalize to a sequence completion task, which more closely resembles the next-token prediction process in LLM pre-training. Together, our findings reveal how information retrieval demands, model architecture, and structural attention dynamics during model training can jointly produce positional bias observed in LLMs.

cs.LG↗

arXiv2025

RAG-IGBench: Innovative Evaluation for RAG-based Interleaved Generation in Open-domain Question Answering

In real-world scenarios, providing user queries with visually enhanced responses can considerably benefit understanding and memory, underscoring the great value of interleaved image-text generation. Despite recent progress, like the visual autoregressive model that unifies text and image processing in a single transformer architecture, generating high-quality interleaved content remains challenging. Moreover, evaluations of these interleaved sequences largely remain underexplored, with existing benchmarks often limited by unimodal metrics that inadequately assess the intricacies of combined image-text outputs. To address these issues, we present RAG-IGBench, a thorough benchmark designed specifically to evaluate the task of Interleaved Generation based on Retrieval-Augmented Generation (RAG-IG) in open-domain question answering. RAG-IG integrates multimodal large language models (MLLMs) with retrieval mechanisms, enabling the models to access external image-text information for generating coherent multimodal content. Distinct from previous datasets, RAG-IGBench draws on the latest publicly available content from social platforms and introduces innovative evaluation metrics that measure the quality of text and images, as well as their consistency. Through extensive experiments with state-of-the-art MLLMs (both open-source and proprietary) on RAG-IGBench, we provide an in-depth analysis examining the capabilities and limitations of these models. Additionally, we validate our evaluation metrics by demonstrating their high correlation with human assessments. Models fine-tuned on RAG-IGBench's training set exhibit improved performance across multiple benchmarks, confirming both the quality and practical utility of our dataset. Our benchmark is available at https://github.com/USTC-StarTeam/RAG-IGBench.

cs.IR↗

arXiv2025

Universality and kernel-adaptive training for classically trained, quantum-deployed generative models

The instantaneous quantum polynomial (IQP) quantum circuit Born machine (QCBM) has been proposed as a promising quantum generative model over bitstrings. Recent works have shown that the training of IQP-QCBM is classically tractable w.r.t. the so-called Gaussian kernel maximum mean discrepancy (MMD) loss function, while maintaining the potential of a quantum advantage for sampling itself. Nonetheless, the model has a number of aspects where improvements would be important for more general utility: (1) the basic model is known to be not universal - i.e. it is not capable of representing arbitrary distributions, and it was not known whether it is possible to achieve universality by adding hidden (ancillary) qubits; (2) a fixed Gaussian kernel used in the MMD loss can cause training issues, e.g., vanishing gradients. In this paper, we resolve the first question and make decisive strides on the second. We prove that for an $n$-qubit IQP generator, adding $n + 1$ hidden qubits makes the model universal. For the latter, we propose a kernel-adaptive training method, where the kernel is adversarially trained. We show that in the kernel-adaptive method, the convergence of the MMD value implies weak convergence in distribution of the generator. We also analytically analyze the limitations of the MMD-based training method. Finally, we verify the performance benefits on the dataset crafted to spotlight improvements by the suggested method. The results show that kernel-adaptive training outperforms a fixed Gaussian kernel in total variation distance, and the gap increases with the dataset dimensionality. These modifications and analyses shed light on the limits and potential of these new quantum generative methods, which could offer the first truly scalable insights in the comparative capacities of classical versus quantum models, even without access to scalable quantum computers.

quant-ph↗

arXiv2025

BEACON: Bayesian Optimal Stopping for Efficient LLM Sampling

Sampling multiple responses is a common way to improve LLM output quality, but it comes at the cost of additional computation. The key challenge is deciding when to stop generating new samples to balance accuracy gains against efficiency. To address this, we introduce BEACON (Bayesian Efficient Adaptive Criterion for Optimal N-stopping), a principled adaptive sampling framework grounded in Sequential Search with Bayesian Learning. BEACON sequentially generates responses from the policy LLM, updates posterior belief over reward distributions in real time without further training, and determines when to stop by weighing expected gains against computational cost. Sampling terminates once the marginal utility of further exploration no longer justifies the expense. We establish both theoretical optimality guarantees and practical tractability, and show empirically that BEACON reduces average sampling by up to 80% while maintaining response quality. We further demonstrate BEACON's utility for cost-efficient preference data generation and outline practical extensions, offering actionable insights for future researchers.

cs.LG↗

arXiv2025

The influence of the mean anomaly on the dynamical quantities of binary black hole mergers in eccentric orbits

In studies of binary black hole (BBH) mergers in eccentric orbits, the mean anomaly, traditionally regarded as less significant than eccentricity, has been thought to encode only the orbital phase, leading to the assumption that it exerts minimal influence on the dynamics of eccentric mergers. In a previous investigation, we identified consistent oscillations in dynamical quantities peak luminosity $L_{\text{peak}}$, remnant mass $M_{\text{rem}}$, spin $α_{\text{rem}}$, and recoil velocity $V_{\text{rem}}$ in relation to the initial eccentricity $e_0$. These oscillations are associated with integer orbital cycles within a phenomenological framework. In this paper, we aim to explore the underlying physical nature of these oscillations through gravitational waveforms. Our examination of remnant mass and spin reveals that while the initial ADM mass $M_{\mathrm{ADM}}$ and orbital angular momentum $L_0$ exhibit gradual variations with $e_0$, the radiated energy $E_{\text{rad}}$ and angular momentum $L_{\text{rad}}$ display oscillatory patterns akin to those observed in $M_{\text{rem}}$ and $α_{\text{rem}}$. By decomposing the waveforms into three distinct phases inspiral, late inspiral to merger, and ringdown, we demonstrate that these oscillations persist across all phases, suggesting a common origin. Through a comparative analysis of $E_{\text{rad}}$ and $L_{\text{rad}}$ derived from numerical relativity (NR), post-Newtonian (PN) waveforms, and orbital-averaged PN fluxes during the inspiral phase, we identify the initial mean anomaly $l_0$ as the source of the observed oscillations. ...

gr-qc↗

arXiv2025

Testing $n_s=1$ in light of the latest ACT and SPT data

It is commonly recognized that the primordial scalar spectral index $n_s$ is approximately $0.96-0.975$, depending on the dataset. However, this view is being completely altered by the early dark energy (EDE) resolutions of the Hubble tension, known as the most prominent tension the standard $Λ$CDM model is suffering from. In corresponding models with pre-recombination EDE, resolving the Hubble tension (i.e., achieving $H_0\sim 73$km/s/Mpc) must be accompanied by a shift of $n_s$ towards unity to maintain consistency with the cosmological data, which thus implies a scale invariant Harrison-Zel'dovich spectrum with $n_s=1$ $(|n_s-1|\simeq {\cal O}(0.001))$. In this work, we strengthen and reconfirm this result with the latest ground-based CMB data from ACT DR6 and SPT-3G D1, the precise measurements at high multipoles beyond the Planck angular resolution and sensitivity. Our work again highlights the importance of re-examining our understanding on the very early Universe within the broader context of cosmological tensions.

astro-ph.CO↗

arXiv2025

Taming Latency-Memory Trade-Off in MoE-Based LLM Serving via Fine-Grained Expert Offloading

Large Language Models (LLMs) have gained immense success in revolutionizing various applications, including content generation, search and recommendation, and AI-assisted operation. To reduce high training costs, Mixture-of-Experts (MoE) architecture has become a popular backbone for modern LLMs. However, despite the benefits, serving MoE-based LLMs experience severe memory inefficiency due to sparsely activated experts. Recent studies propose to offload inactive experts from GPU memory to CPU memory to improve the serving efficiency of MoE models. However, they either incur high inference latency or high model memory footprints due to coarse-grained designs. To tame the latency-memory trade-off in MoE serving, we present FineMoE, a fine-grained expert offloading system for MoE serving that achieves low inference latency with memory efficiency. We design FineMoE to extract fine-grained expert selection patterns from MoE models and semantic hints from input prompts to efficiently guide expert prefetching, caching, and offloading decisions. FineMoE is prototyped on top of HuggingFace Transformers and deployed on a six-GPU testbed. Experiments with open-source MoE models and real-world workloads show that FineMoE reduces inference latency by 47% and improves expert hit rate by 39% over state-of-the-art solutions.

cs.LG↗

arXiv2025

Towards Convexity in Anomaly Detection: A New Formulation of SSLM with Unique Optimal Solutions

An unsolved issue in widely used methods such as Support Vector Data Description (SVDD) and Small Sphere and Large Margin SVM (SSLM) for anomaly detection is their nonconvexity, which hampers the analysis of optimal solutions in a manner similar to SVMs and limits their applicability in large-scale scenarios. In this paper, we introduce a novel convex SSLM formulation which has been demonstrated to revert to a convex quadratic programming problem for hyperparameter values of interest. Leveraging the convexity of our method, we derive numerous results that are unattainable with traditional nonconvex approaches. We conduct a thorough analysis of how hyperparameters influence the optimal solution, pointing out scenarios where optimal solutions can be trivially found and identifying instances of ill-posedness. Most notably, we establish connections between our method and traditional approaches, providing a clear determination of when the optimal solution is unique--a task unachievable with traditional nonconvex methods. We also derive the nu-property to elucidate the interactions between hyperparameters and the fractions of support vectors and margin errors in both positive and negative classes.

cs.LG↗

arXiv2025

MindVL: Towards Efficient and Effective Training of Multimodal Large Language Models on Ascend NPUs

We propose MindVL, a multimodal large language model (MLLMs) trained on Ascend NPUs. The training of state-of-the-art MLLMs is often confined to a limited set of hardware platforms and relies heavily on massive, undisclosed data recipes, which hinders reproducibility and open research. To change the common perception that Ascend hardware is unsuitable for efficient full-stage MLLM training, we introduce MindSpeed-MLLM, a highly efficient training framework that supports stable and high-performance training of large-scale Dense and Mixture-of-Experts (MoE) models on Ascend hardware. Based on this, we provide a systematic and open description of the data production methods and mixing strategies for all training stages. Furthermore, we present MindVL, a data-efficient multimodal large language model trained end-to-end on Ascend NPUs. In addition, we find that averaging weights from checkpoints trained with different sequence lengths is particularly effective and yields further gains when combined with test-time resolution search. Our experiments demonstrate superior data efficiency: MindVL-8B matches the performance of Qwen2.5VL-7B using only 10\% of its training data, while our MoE model, MindVL-671B-A37B, matches Qwen2.5VL-72B using only 3\% of the Qwen2.5VL training data, and achieves comparable performance with other leading multimodal MoE models. Our work provides the community with a valuable hardware alternative, open data recipes, and effective performance-enhancing techniques.

cs.CV↗

arXiv2025

Electromagnetically driven, environmentally adaptive, and functionally switchable hydrodynamic devices

Metamaterials provide exceptional control over physical phenomena, enabling many disruptive technologies. However, researches in hydrodynamic meta-devices have mainly used intrusive methods to manipulate material structures, limited by material properties and specific environmental conditions. Each design serves a single function, reducing versatility. This study introduces a meta-hydrodynamics theory using applied force fields to avoid physical contact with the fluid and eliminate the need for inhomogeneous and anisotropic metamaterials, allowing continuous switching between cloaking, shielding, and Venturi amplification. The force field operates independently of the fluid's physical properties, making it adaptable to various fluids and environmental conditions. We derive volumetric force distributions for hydrodynamic devices based on fluid properties and forces equivalence, using the integral median theorem to homogenize these forces for practical applications. The effectiveness of the proposed hydrodynamic devices is validated through numerical simulations and quantitative analyses. By utilizing the electromagnetic forces produced by the interaction between a conducting fluid and an electromagnetic field, we experimentally verified the validity of our theoretical simulations. Our research offers different insights into hydrodynamic meta-devices design, enhancing practical applications and opening avenues for innovative flow manipulation.

physics.flu-dyn↗

arXiv2025

Private Information Retrieval over Graphs

The problem of PIR in graph-based replication systems has received significant attention in recent years. A systematic study was conducted by Sadeh, Gu, and Tamo, where each file is replicated across two servers and the storage topology is modeled by a graph. The PIR capacity of a graph $G$, denoted by $\mathcal{C}(G)$, is defined as the supremum of retrieval rates achievable by schemes that preserve user privacy, with the rate measured as the ratio between the file size and the total number of bits downloaded. This paper makes the following key contributions. (1) The complete graph $K_N$ has emerged as a central benchmark in the study of PIR over graphs. The asymptotic gap between the upper and lower bounds for $\mathcal{C}(K_N)$ was previously 2 and was only recently reduced to $5/3$. We shrink this gap to $1.0444$, bringing it close to resolution. More precisely, (i) Sadeh, Gu, and Tamo proved that $\mathcal{C}(K_N)\le 2/(N+1)$ and conjectured this bound to be tight. We refute this conjecture by establishing the strictly stronger bound $\mathcal{C}(K_N) \le \frac{1.3922}{N}.$ We also improve the upper bound for the balanced complete bipartite graph $\mathcal{C}(K_{N/2,N/2})$. (ii) The first lower bound on $\mathcal{C}(K_N)$ was $(1+o(1))/N$, which was recently sharpened to $(6/5+o(1))/N$. We provide explicit, systematic constructions that further improve this bound, proving $\mathcal{C}(K_N)\ge(4/3-o(1))/N,$ which in particular implies $\mathcal{C}(G) \ge (4/3-o(1))/|G|$ for every graph $G$. (2) We establish a conceptual bridge between deterministic and probabilistic PIR schemes on graphs. This connection has significant implications for reducing the required subpacketization in practical implementations and is of independent interest. We also design a general probabilistic PIR scheme that performs particularly well on sparse graphs.

cs.IT↗

arXiv2025

TriMat: Context-aware Recommendation by Tri-Matrix Factorization

Search engine is the symbolic technology of Web 2.0, and many people used to believe recommender systems is the new frontier of Web 3.0. In the past 10 years, with the advent of TikTok and similar apps, recommender systems has materialized the vision of the machine learning pioneers. However, many research topics of the field remain unfixed until today. One such topic is CARS (Context-aware Recommender Systems) , which is largely a theoretical topic without much advance in real-world applications. In this paper, we utilize tri-matrix factorization technique to incorporate contextual information into our matrix factorization framework, and prove that our technique is effective in improving both the accuracy and fairness metrics in our experiments.

cs.IR↗

arXiv2025

When Does Multimodality Lead to Better Time Series Forecasting?

Recently, there has been growing interest in incorporating textual information into foundation models for time series forecasting. However, it remains unclear whether and under what conditions such multimodal integration consistently yields gains. We systematically investigate these questions across a diverse benchmark of 16 forecasting tasks spanning 7 domains, including health, environment, and economics. We evaluate two popular multimodal forecasting paradigms: aligning-based methods, which align time series and text representations; and prompting-based methods, which directly prompt large language models for forecasting. Our findings reveal that the benefits of multimodality are highly condition-dependent. While we confirm reported gains in some settings, these improvements are not universal across datasets or models. To move beyond empirical observations, we disentangle the effects of model architectural properties and data characteristics, drawing data-agnostic insights that generalize across domains. Our findings highlight that on the modeling side, incorporating text information is most helpful given (1) high-capacity text models, (2) comparatively weaker time series models, and (3) appropriate aligning strategies. On the data side, performance gains are more likely when (4) sufficient training data is available and (5) the text offers complementary predictive signal beyond what is already captured from the time series alone. Our study offers a rigorous, quantitative foundation for understanding when multimodality can be expected to aid forecasting tasks, and reveals that its benefits are neither universal nor always aligned with intuition.

cs.CL↗

arXiv2025

Demonstration of quantum error detection in a silicon quantum processor

Quantum error detection is essential in realizing large-scale universal quantum computation, especially for quantum error correction (QEC). However, key elements for FTQC have yet to be realized in silicon qubits. Here, we demonstrate quantum error detection on a donor-based silicon quantum processor comprising four-nuclear spin qubits and one electron spin as an auxiliary qubit. The entanglement capability of this system is validated through the establishment of two-qubit Bell state entanglement between the nuclear spins and the generation of a four-qubit Greenberger-Horne-Zeilinger (GHZ) state, achieving a GHZ state fidelity of 88.5(2.3)%. Furthermore, by executing a four-qubit error detection circuit with the stabilizers, we successfully detect arbitrary single-qubit errors. The encoded Bell state entanglement information is recovered by performing the Pauli-frame update (PFU) via postprocessing. Based on the detected errors, we identify strongly biased noise in our system. Our results mark a significant advance toward FTQC in silicon spin qubits.

quant-ph↗

arXiv2025

BioVessel-Net and RetinaMix: Unsupervised Retinal Vessel Segmentation from OCTA Images

Structural changes in retinal blood vessels are critical biomarkers for the onset and progression of glaucoma and other ocular diseases. However, current vessel segmentation approaches largely rely on supervised learning and extensive manual annotations, which are costly, error-prone, and difficult to obtain in optical coherence tomography angiography. Here we present BioVessel-Net, an unsupervised generative framework that integrates vessel biostatistics with adversarial refinement and a radius-guided segmentation strategy. Unlike pixel-based methods, BioVessel-Net directly models vascular structures with biostatistical coherence, achieving accurate and explainable vessel extraction without labeled data or high-performance computing. To support training and evaluation, we introduce RetinaMix, a new benchmark dataset of 2D and 3D OCTA images with high-resolution vessel details from diverse populations. Experimental results demonstrate that BioVessel-Net achieves near-perfect segmentation accuracy across RetinaMix and existing datasets, substantially outperforming state-of-the-art supervised and semi-supervised methods. Together, BioVessel-Net and RetinaMix provide a label-free, computationally efficient, and clinically interpretable solution for retinal vessel analysis, with broad potential for glaucoma monitoring, blood flow modeling, and progression prediction. Code and dataset are available: https://github.com/VikiXie/SatMar8.

cs.CV↗

arXiv2025

Hint of $r\simeq 0.01$ after DESI DR2 ?

In the report by BICEP/Keck collaborations, the tensor-to-scalar ratio is $r_{0.05}<0.036$ (95\% C.L.). However, recent datasets have preferred the evolving dark energy, which thus have significantly shifted the bestfit values of standard $Λ$CDM cosmological parameters. In this paper, we perform the joint analysis of BICEP/Keck cosmic microwave background (CMB) B-mode data, latest DESI DR2 baryon acoustic oscillations and supernova data, combined with Planck PR3 and PR4 CMB data respectively, and find $r_{0.05}=0.0159^{+0.0057}_{-0.014}$ and $r_{0.05}=0.0164^{+0.0063}_{-0.014}$. The constraints on $r$ are further tightened compared to the result of BICEP/Keck collaborations. Though there might be still systematic uncertainties in B-mode measurements due to the foreground contamination, our work is to not say what the value of $r$ is, but present the state-of-the-art constraints on $r$ and emphasize that the detection for $r$ depends potentially on our insight into the dark universe, highlighting the important role of cosmological surveys in comprehending our very early universe.

astro-ph.CO↗