arXiv subjects

Yuqi Wang

Publications and source records attributed to Yuqi Wang.

At least 55 records · Page 3Linked to original sources

arXiv2025

K2-Think: A Parameter-Efficient Reasoning System

K2-Think is a reasoning system that achieves state-of-the-art performance with a 32B parameter model, matching or surpassing much larger models like GPT-OSS 120B and DeepSeek v3.1. Built on the Qwen2.5 base model, our system shows that smaller models can compete at the highest levels by combining advanced post-training and test-time computation techniques. The approach is based on six key technical pillars: Long Chain-of-thought Supervised Finetuning, Reinforcement Learning with Verifiable Rewards (RLVR), Agentic planning prior to reasoning, Test-time Scaling, Speculative Decoding, and Inference-optimized Hardware, all using publicly available open-source datasets. K2-Think excels in mathematical reasoning, achieving state-of-the-art scores on public benchmarks for open-source models, while also performing strongly in other areas such as Code and Science. Our results confirm that a more parameter-efficient model like K2-Think 32B can compete with state-of-the-art systems through an integrated post-training recipe that includes long chain-of-thought training and strategic inference-time enhancements, making open-source reasoning systems more accessible and affordable. K2-Think is freely available at k2think.ai, offering best-in-class inference speeds of over 2,000 tokens per second per request via the Cerebras Wafer-Scale Engine.

cs.LG↗

arXiv2025

A unified diagrammatic formulation of single-reference and multi-reference random phase approximations: the particle-hole and particle-particle channels

A diagrammatic multi-reference generalization of many-body perturbation theory was recently introduced [J. Phys. Chem. Lett., 2025, 16, 3047]. This framework allows us to extend single-reference (SR) Green's function methods defined at the diagrammatic level naturally into multi-reference case, as previously exemplified by the formulation of multi-reference direct random phase approximation (MR-dRPA) and the multi-reference second-order screened exchange approximation (MR-SOSEX). In this work, we further elaborate this framework and use it to develop MR generalizations of two other RPA variants, namely, particle-hole (ph) RPA with exchange (MR-RPAx) and particle-particle RPA (MR-ppRPA). We define these two MR generalizations by infinite order resummations of the generalized `ring' and `ladder' diagrams with antisymmetrized interaction vertices, respectively, which incorporate the contributions from the active-space connected two-body Green's functions. As for MR-dRPA, we derive unified sets of equations that hold at both SR and MR levels for RPAx and ppRPA, respectively. We perform numerical studies of prototypical systems using the three MR-RPA methods and carry out a perturbative analysis to gain a deeper understanding of their behaviors. We find that error cancellation between the second and third orders is a key factor for both SR-RPA and MR-RPA. In addition, we observe that MR-phRPA (MR-dRPA and MR-RPAx) and MR-ppRPA tend to overestimate and underestimate correlation energies, respectively, suggesting that a better accuracy can be achieved by further combining these two channels in the future.

quant-ph↗

arXiv2025

Force-IMU Fusion-Based Sensing Acupuncture Needle and Quantitative Analysis System for Acupuncture Manipulations

Acupuncture, one of the key therapeutic methods in Traditional Chinese Medicine (TCM), has been widely adopted in various clinical fields. Quantitative research on acupuncture manipulation parameters is critical to achieve standardized techniques. However, quantitative mechanical detection of acupuncture parameters remains limited. This study establishes a kinematic and dynamic model of acupuncture, identifying key parameters such as lifting-thrusting force, acceleration, velocity, displacement, as well as twirling-rotating angular velocity and angle. To measure these critical parameters, we propose a quantitative system comprising a sensing needle equipped with a force sensor and an inertial measurement unit (IMU), as well as an external camera module to capture image information. By fusing visual and IMU data, we accurately identify the stationary or motion states of the needle, enabling segmented computation of lifting-thrusting velocity and displacement. The experimental results demonstrate that the sensing needle achieves comprehensive detection with high precision, featuring a nonlinearity error of 0.45% in force measurement and an RMSE of 1.2 mm in displacement. The extracted parameters provide an objective description of the operational characteristics and motion patterns of the four basic acupuncture manipulations. These findings provide valuable tools and methods for research in acupuncture standardization.

eess.SY↗

arXiv2025

Teacher Motion Priors: Enhancing Robot Locomotion over Challenging Terrain

Achieving robust locomotion on complex terrains remains a challenge due to high dimensional control and environmental uncertainties. This paper introduces a teacher prior framework based on the teacher student paradigm, integrating imitation and auxiliary task learning to improve learning efficiency and generalization. Unlike traditional paradigms that strongly rely on encoder-based state embeddings, our framework decouples the network design, simplifying the policy network and deployment. A high performance teacher policy is first trained using privileged information to acquire generalizable motion skills. The teacher's motion distribution is transferred to the student policy, which relies only on noisy proprioceptive data, via a generative adversarial mechanism to mitigate performance degradation caused by distributional shifts. Additionally, auxiliary task learning enhances the student policy's feature representation, speeding up convergence and improving adaptability to varying terrains. The framework is validated on a humanoid robot, showing a great improvement in locomotion stability on dynamic terrains and significant reductions in development costs. This work provides a practical solution for deploying robust locomotion strategies in humanoid robots.

cs.RO↗

arXiv2025

Unified Vision-Language-Action Model

Vision-language-action models (VLAs) have garnered significant attention for their potential in advancing robotic manipulation. However, previous approaches predominantly rely on the general comprehension capabilities of vision-language models (VLMs) to generate action signals, often overlooking the rich temporal and causal structure embedded in visual observations. In this paper, we present UniVLA, a unified and native multimodal VLA model that autoregressively models vision, language, and action signals as discrete token sequences. This formulation enables flexible multimodal tasks learning, particularly from large-scale video data. By incorporating world modeling during post-training, UniVLA captures causal dynamics from videos, facilitating effective transfer to downstream policy learning--especially for long-horizon tasks. Our approach sets new state-of-the-art results across several widely used simulation benchmarks, including CALVIN, LIBERO, and Simplenv-Bridge, significantly surpassing previous methods. For example, UniVLA achieves 95.5% average success rate on LIBERO benchmark, surpassing pi0-FAST's 85.5%. We further demonstrate its broad applicability on real-world ALOHA manipulation and autonomous driving.

cs.CV↗

arXiv2025

Revisiting Reinforcement Learning for LLM Reasoning from A Cross-Domain Perspective

Reinforcement learning (RL) has emerged as a promising approach to improve large language model (LLM) reasoning, yet most open efforts focus narrowly on math and code, limiting our understanding of its broader applicability to general reasoning. A key challenge lies in the lack of reliable, scalable RL reward signals across diverse reasoning domains. We introduce Guru, a curated RL reasoning corpus of 92K verifiable examples spanning six reasoning domains--Math, Code, Science, Logic, Simulation, and Tabular--each built through domain-specific reward design, deduplication, and filtering to ensure reliability and effectiveness for RL training. Based on Guru, we systematically revisit established findings in RL for LLM reasoning and observe significant variation across domains. For example, while prior work suggests that RL primarily elicits existing knowledge from pretrained models, our results reveal a more nuanced pattern: domains frequently seen during pretraining (Math, Code, Science) easily benefit from cross-domain RL training, while domains with limited pretraining exposure (Logic, Simulation, and Tabular) require in-domain training to achieve meaningful performance gains, suggesting that RL is likely to facilitate genuine skill acquisition. Finally, we present Guru-7B and Guru-32B, two models that achieve state-of-the-art performance among open models RL-trained with publicly available data, outperforming best baselines by 7.9% and 6.7% on our 17-task evaluation suite across six reasoning domains. We also show that our models effectively improve the Pass@k performance of their base models, particularly on complex tasks less likely to appear in pretraining data. We release data, models, training and evaluation code to facilitate general-purpose reasoning at: https://github.com/LLM360/Reasoning360

cs.LG↗

arXiv2025

An efficient GPU-accelerated adaptive mesh refinement framework for high-fidelity compressible reactive flows modeling

This paper presents a heterogeneous adaptive mesh refinement (AMR) framework for efficient simulation of moderately stiff reactive problems. This framework features an elaborate subcycling-in-time algorithm along with a specialized refluxing algorithm, all unified in a highly parallel codebase. We have also developed a low-storage variant of explicit chemical integrators by optimizing the register usage of GPU, achieving respectively 6x and 3x times speedups as compared to the implicit and standard explicit methods with comparable order of accuracy. A suite of benchmarks have confirmed the framework's fidelity for both non-reactive and reactive simulations with/without AMR. By leveraging our parallelization strategy that is developed on AMReX, we have demonstrated remarkable speedups on various problems on a NVIDIA V100 GPU than using a Intel i9 CPU within the same codebase; in problems with complex physics and spatiotemporally distributed stiffness such as the hydrogen detonation propagation, we have achieved an overall 6.49x acceleration ratio. The computation scalability of the framework is also validated through the weak scaling test, demonstrating excellent parallel efficiency across multiple GPU nodes. At last, a practical application of this GPU-accelerated SAMR framework to large-scale direct numerical simulations is demonstrated by successful simulation of the three-dimensional reactive shock-bubble interaction problem; we have saved significant computational costs while maintaining the comparable accuracy, as compared to a prior uniform DNS study performed on CPUs.

physics.comp-ph↗

arXiv2025

A stochastic force model for a finite-size spherical particle in turbulence

Predicting particle-laden flows requires accurate fluid force models. However, a reliable particle force model for finite-size particles in turbulent flows remains lacking. In the present work, a fluid force model for a finite-size spherical particle in turbulence is developed by simulating turbulent flow past a fixed spherical particle using particle-resolved direct numerical simulation (PRDNS). Our simulation demonstrates that turbulence increases the mean drag force of the particle, which is consistent with previous studies. By correlating the DNS data as functions of the Reynolds number of particles, the ratio of the particle-to-turbulence scale, and the intensity ratio of the turbulence, an empirical correlation for the mean drag force is obtained. Furthermore, we find that the fluctuations of both the drag and lateral forces follow the Gaussian distribution. Consequently, the temporal variations of the fluctuating drag and lateral forces are modeled using a stochastic Langevin equation. Empirical correlations of the fluctuation intensities and time scales involved in the stochastic model are also determined from the DNS data. Finally, we simulate the movement of a finite-size particle in turbulence and the dispersion of particles in a turbulent channel flow to validate the proposed model. The proposed fluid force model requires the mean flow velocity, the kinetic energy of the turbulence, and the dissipation rate of the turbulence as inputs, which makes it well suited for combination with the Reynolds-averaged Navier-Stokes (RANS) approach.

physics.flu-dyn↗

arXiv2025

From Temporal to Spatial: Designing Spatialized Interactions with Segmented-audios in Immersive Environments for Active Engagement with Performing Arts Intangible Cultural Heritage

Performance artforms like Peking opera face transmission challenges due to the extensive passive listening required to understand their nuance. To create engaging forms of experiencing auditory Intangible Cultural Heritage (ICH), we designed a spatial interaction-based segmented-audio (SISA) Virtual Reality system that transforms passive ICH experiences into active ones. We undertook: (1) a co-design workshop with seven stakeholders to establish design requirements, (2) prototyping with five participants to validate design elements, and (3) user testing with 16 participants exploring Peking Opera. We designed transformations of temporal music into spatial interactions by cutting sounds into short audio segments, applying t-SNE algorithm to cluster audio segments spatially. Users navigate through these sounds by their similarity in audio property. Analysis revealed two distinct interaction patterns (Progressive and Adaptive), and demonstrated SISA's efficacy in facilitating active auditory ICH engagement. Our work illuminates the design process for enriching traditional performance artform using spatially-tuned forms of listening.

cs.HC↗

arXiv2025

Dark Superradiance in Cavity-Coupled Polar Molecular Bose-Einstein Condensates

We propose an experimental scheme to realize phase transition from {\it dark superradiance} to conventional superradiance in a microwave cavity coupled to polar molecules. The competition between cavity-mediated infinite-range repulsions and finite-range attractive dipolar interactions stabilizes a variety of exotic quantum phases, including vortex, vortex anti-vortex pairs, and superradiant phase, all emerging without external driving fields. In vortex phase associated with {\it dark superradiance}, cavity remains in vacuum state while profoundly reshaping the condensate's ground-state wave functions. In particular, the spin configuration locally parallel but globally anti-parallel is a direct consequence of competing for two nonlocal interactions. Beyond Dicke paradigm, dipolar dressing of condensate enables access to an unexplored regime of repulsion-dominated superradiance. A Bogoliubov analysis of low-energy excitation spectrum confirms that the condensate remains stable, avoiding roton-maxon induced collapse even in strongly dipolar regime.

cond-mat.quant-gas↗

arXiv2025

End-to-End Driving with Online Trajectory Evaluation via BEV World Model

End-to-end autonomous driving has achieved remarkable progress by integrating perception, prediction, and planning into a fully differentiable framework. Yet, to fully realize its potential, an effective online trajectory evaluation is indispensable to ensure safety. By forecasting the future outcomes of a given trajectory, trajectory evaluation becomes much more effective. This goal can be achieved by employing a world model to capture environmental dynamics and predict future states. Therefore, we propose an end-to-end driving framework WoTE, which leverages a BEV World model to predict future BEV states for Trajectory Evaluation. The proposed BEV world model is latency-efficient compared to image-level world models and can be seamlessly supervised using off-the-shelf BEV-space traffic simulators. We validate our framework on both the NAVSIM benchmark and the closed-loop Bench2Drive benchmark based on the CARLA simulator, achieving state-of-the-art performance. Code is released at https://github.com/liyingyanUCAS/WoTE.

cs.CV↗

arXiv2025

Generalized many-body perturbation theory for the electron correlation energy: multi-reference random phase approximation via diagrammatic resummation

Many-body perturbation theory (MBPT) based on Green's functions and Feynman diagrams provides a fundamental theoretical framework for various \emph{ab initio} computational approaches in molecular and materials science, including the random phase approximation (RPA) and $GW$ approximation. Unfortunately, this perturbation expansion often fails in systems with strong multi-reference characters. Extending diagrammatic MBPT to the multi-reference case is highly nontrivial and remains largely unexplored, primarily due to the breakdown of Wick's theorem. In this work, we develop a diagrammatic multi-reference generalization of MBPT for computing correlation energies of strongly correlated systems, by using the cumulant expansion of many-body Green's function in place of Wick's theorem. This theoretical framework bridges the gap between MBPT in condensed matter physics and multi-reference perturbation theories (MRPT) in quantum chemistry, which had been almost exclusively formulated within time-independent wavefunction frameworks prior to this work. Our formulation enables the explicit incorporation of strong correlation effects from the outset as in MRPT, while treating residual weak interactions through a generalized diagrammatic perturbation expansion as in MBPT. As a concrete demonstration, we formulate a multi-reference (MR) extension of the standard single-reference (SR) RPA by systematically resumming generalized ring diagrams, which naturally leads to a unified set of equations applicable to both SR and MR cases. Benchmark calculations on prototypical molecular systems reveal that MR-RPA successfully resolves the well-known failure of SR-RPA in strongly correlated systems. This theoretical advancement paves the way for advancing \emph{ab initio} computational methods through diagrammatic resummation techniques in future.

quant-ph↗

arXiv2025

Enhancing End-to-End Autonomous Driving with Latent World Model

In autonomous driving, end-to-end planners directly utilize raw sensor data, enabling them to extract richer scene features and reduce information loss compared to traditional planners. This raises a crucial research question: how can we develop better scene feature representations to fully leverage sensor data in end-to-end driving? Self-supervised learning methods show great success in learning rich feature representations in NLP and computer vision. Inspired by this, we propose a novel self-supervised learning approach using the LAtent World model (LAW) for end-to-end driving. LAW predicts future scene features based on current features and ego trajectories. This self-supervised task can be seamlessly integrated into perception-free and perception-based frameworks, improving scene feature learning and optimizing trajectory prediction. LAW achieves state-of-the-art performance across multiple benchmarks, including real-world open-loop benchmark nuScenes, NAVSIM, and simulator-based closed-loop benchmark CARLA. The code is released at https://github.com/BraveGroup/LAW.

cs.CV↗

arXiv2025

Argus: Benchmarking and Enhancing Vision-Language Models for 3D Radiology Report Generation

Automatic radiology report generation holds significant potential to streamline the labor-intensive process of report writing by radiologists, particularly for 3D radiographs such as CT scans. While CT scans are critical for clinical diagnostics, they remain less explored compared to 2D radiographs. To date, there has been no comprehensive benchmark for 3D radiograph report generation (3DRRG), nor sufficient investigation into the optimal training strategies for Vision Language Models (VLMs) in this context, particularly with respect to vision encoder choices, visual token compression, and model scaling. In this work, we make three key contributions. We curate **CT-3DRRG**, the largest **publicly** available 3D CT-report dataset, establishing a robust and diverse benchmark for evaluating VLM performance on 3DRRG. Furthermore, we propose a comprehensive training recipe for building high-performing VLMs for 3DRRG, exploring key factors such as vision encoder pretraining strategies, visual token compression, and the impact of data & model scale. Guided by these findings, we introduce **Argus**, a state-of-the-art family of VLMs that achieve superior performance across different model sizes and input 3D medical image resolutions, efficiently processing high-resolution 3D images up to $512 \times 512 \times 256$[^1].

cs.CV↗

arXiv2025

LLM360 K2: Building a 65B 360-Open-Source Large Language Model from Scratch

We detail the training of the LLM360 K2-65B model, scaling up our 360-degree OPEN SOURCE approach to the largest and most powerful models under project LLM360. While open-source LLMs continue to advance, the answer to "How are the largest LLMs trained?" remains unclear within the community. The implementation details for such high-capacity models are often protected due to business considerations associated with their high cost. This lack of transparency prevents LLM researchers from leveraging valuable insights from prior experience, e.g., "What are the best practices for addressing loss spikes?" The LLM360 K2 project addresses this gap by providing full transparency and access to resources accumulated during the training of LLMs at the largest scale. This report highlights key elements of the K2 project, including our first model, K2 DIAMOND, a 65 billion-parameter LLM that surpasses LLaMA-65B and rivals LLaMA2-70B, while requiring fewer FLOPs and tokens. We detail the implementation steps and present a longitudinal analysis of K2 DIAMOND's capabilities throughout its training process. We also outline ongoing projects such as TXT360, setting the stage for future models in the series. By offering previously unavailable resources, the K2 project also resonates with the 360-degree OPEN SOURCE principles of transparency, reproducibility, and accessibility, which we believe are vital in the era of resource-intensive AI research.

cs.LG↗

arXiv2025

Atom-Molecule Superradiance and Entanglement with Cavity-Mediated Three-Body Interactions

Ultracold atoms coupled to optical cavities offer a powerful platform for studying strongly correlated many-body physics. Here, we propose an experimental scheme for creating biatomic molecules via cavity-enhanced photoassociation from an atomic condensate. This setup realizes long-range three-body interactions mediated by tripartite cavity-atom-molecule coupling. Beyond a critical pump strength, a self-organized square lattice phase for molecular condensate emerges, resulting in hybrid atom-molecule superradiance with spontaneous $U(1)$ symmetry breaking. Distinct from previously observed ultracold bosonic (fermionic) atomic superradiance, our findings demonstrate bosonic enhancement characterized by a cubic scaling of steady-state photon number with total atom number. Additionally, strong photon-matter entanglement is shown to effectively characterize superradiant quantum phase transition. Our findings deepen the understanding of quantum superchemistry and exotic many-body nonequilibrium dynamics in cavity-coupled quantum gases.

cond-mat.quant-gas↗

arXiv2024

DrivingGPT: Unifying Driving World Modeling and Planning with Multi-modal Autoregressive Transformers

World model-based searching and planning are widely recognized as a promising path toward human-level physical intelligence. However, current driving world models primarily rely on video diffusion models, which specialize in visual generation but lack the flexibility to incorporate other modalities like action. In contrast, autoregressive transformers have demonstrated exceptional capability in modeling multimodal data. Our work aims to unify both driving model simulation and trajectory planning into a single sequence modeling problem. We introduce a multimodal driving language based on interleaved image and action tokens, and develop DrivingGPT to learn joint world modeling and planning through standard next-token prediction. Our DrivingGPT demonstrates strong performance in both action-conditioned video generation and end-to-end planning, outperforming strong baselines on large-scale nuPlan and NAVSIM benchmarks.

cs.CV↗

arXiv2024

An efficient, adaptive solver for accurate simulation of multicomponent shock-interface problems for thermally perfect species

A second-order-accurate finite volume method, hybridized by blending an extended double-flux algorithm and a traditionally conservative scheme, is developed. In this scheme, hybrid convective fluxes as well as hybrid interpolation techniques are designed to ensure stability and accuracy in the presence of both material interfaces and shocks. Two computationally efficient approaches, extended from the original double-flux model, are presented to eliminate the well-known "pressure oscillation" phenomenon at material interfaces observed with the traditional conservative scheme. Numerous verification simulations confirm that the method is capable of handling multi-dimensional shock-interface problems reliably and efficiently, even in the presence of viscous and reactive terms.

physics.comp-ph↗