arXiv subjects

Yixiao Chen

Publications and source records attributed to Yixiao Chen.

At least 37 records · Page 2Linked to original sources

arXiv2026

Effective MoE-based LLM Compression by Exploiting Heterogeneous Inter-Group Experts Routing Frequency and Information Density

Mixture-of-Experts (MoE) based Large Language Models (LLMs) have achieved superior performance, yet the massive memory overhead caused by storing multiple expert networks severely hinders their practical deployment. Singular Value Decomposition (SVD)-based compression has emerged as a promising post-training technique; however, most existing methods apply uniform rank allocation or rely solely on static weight properties. This overlooks the substantial heterogeneity in expert utilization observed in MoE models, where frequent routing patterns and intrinsic information density vary significantly across experts. In this work, we propose RFID-MoE, an effective framework for MoE compression by exploiting heterogeneous Routing Frequency and Information Density. We first introduce a fused metric that combines expert activation frequency with effective rank to measure expert importance, adaptively allocating higher ranks to critical expert groups under a fixed budget. Moreover, instead of discarding compression residuals, we reconstruct them via a parameter-efficient sparse projection mechanism to recover lost information with minimal parameter overhead. Extensive experiments on representative MoE LLMs (e.g., Qwen3, DeepSeekMoE) across multiple compression ratios demonstrate that RFID-MoE consistently outperforms state-of-the-art methods like MoBE and D2-MoE. Notably, RFID-MoE achieves a perplexity of 16.92 on PTB with the Qwen3-30B model at a 60% compression ratio, reducing perplexity by over 8.0 compared to baselines, and improves zero-shot accuracy on HellaSwag by approximately 8%.

cs.LG↗

arXiv2026

DyMixOp: A Neural Operator Designed from a Complex Dynamics Perspective with Local-Global Mixing for Solving PDEs

A primary challenge in using neural networks to approximate nonlinear dynamical systems governed by partial differential equations (PDEs) lies in recasting these systems into a tractable representation particularly when the dynamics are inherently non-linearizable or require infinite-dimensional spaces for linearization. To address this challenge, we introduce DyMixOp, a novel neural operator framework for PDEs that integrates theoretical insights from complex dynamical systems. Grounded in dynamics-aware priors and inertial manifold theory, DyMixOp projects the original infinite-dimensional PDE dynamics onto a finite-dimensional latent space. This reduction preserves both essential linear structures and dominant nonlinear interactions, thereby establishing a physically interpretable and computationally structured foundation. Central to this approach is the local-global mixing (LGM) transformation, a key architectural innovation inspired by the convective nonlinearity in turbulent flows. By multiplicatively coupling local fine-scale features with global spectral information, LGM effectively captures high-frequency details and complex nonlinear couplings while mitigating the spectral bias that plagues many existing neural operators. The framework is further enhanced by a dynamics-informed architecture that stacks multiple LGM layers in a hybrid configuration, incorporating timescale-adaptive gating and parallel aggregation of intermediate dynamics. This design enables robust approximation of general evolutionary dynamics across diverse physical regimes. Extensive experiments on seven benchmark PDE systems spanning 1D to 3D, elliptic to hyperbolic types demonstrate that DyMixOp achieves state-of-the-art performance on six of them, significantly reducing prediction errors (by up to 94.3% in chaotic regimes) while maintaining computational efficiency and strong scalability.

cs.LG↗

arXiv2025

Ab Initio Melting Properties of Water and Ice from Machine Learning Potentials

Liquid water exhibits several important anomalous properties in the vicinity of the melting temperature ($T_{\mathrm{m}}$) of ice Ih, including a higher density than ice and a density maximum at 4~$^{\circ}$C. Experimentally, an isotope effect on $T_{\mathrm{m}}$ is observed: the melting temperature of H$_2$O is approximately 4~K lower than that of D$_2$O. This difference can only be explained by nuclear quantum effects (NQEs), which can be accurately captured using path integral molecular dynamics (PIMD). Here we run PIMD simulations driven by Deep Potential (DP) models trained on data from density functional theory (DFT) based on SCAN, revPBE0-D3, SCAN0, and revPBE-D3 and a DP model trained on the MB-pol potential. We calculate the \tm of ice, the density discontinuity at melting, and the temperature of density maximum ($T_{\mathrm{dm}}$) of the liquid. We find that the model based on MB-pol agrees well with experiment. The models based on DFT incorrectly predict that NQEs lower $T_{\mathrm{m}}$. For the density discontinuity, SCAN and SCAN0 predict values close to the experimental result, while revPBE-D3 and revPBE0-D3 significantly underestimate it. Additionally, the models based on SCAN and SCAN0 correctly predict that the $T_{\mathrm{dm}}$ is higher than $T_{\mathrm{m}}$, while those based on revPBE-D3 and revPBE0-D3 predict the opposite. We attribute the deviations of the DFT-based models from experiment to the overestimation of hydrogen bond strength. Our results set the stage for more accurate simulations of aqueous systems grounded on DFT.

physics.chem-ph↗

arXiv2025

Assessment of First-Principles Methods in Modeling the Melting Properties of Water

First-principles simulations have played a crucial role in deepening our understanding of the thermodynamic properties of water, and machine learning potentials (MLPs) trained on these first-principles data widen the range of accessible properties. However, the capabilities of different first-principles methods are not yet fully understood due to the lack of systematic benchmarks, the underestimation of the uncertainties introduced by MLPs, and the neglect of nuclear quantum effects (NQEs). Here, we systematically assess first-principles methods by calculating key melting properties using path integral molecular dynamics (PIMD) driven by Deep Potential (DP) models trained on data from density functional theory (DFT) with SCAN, revPBE0-D3, SCAN0 and revPBE-D3 functionals, as well as from the MB-pol potential. We find that MB-pol is in qualitatively good agreement with the experiment in all properties tested, whereas the four DFT functionals incorrectly predict that NQEs increase the melting temperature. SCAN and SCAN0 slightly underestimate the density change between water and ice upon melting, but revPBE-D3 and revPBE0-D3 severely underestimate it. Moreover, SCAN and SCAN0 correctly predict that the maximum liquid density occurs at a temperature higher than the melting point, while revPBE-D3 and revPBE0-D3 predict the opposite behavior. Our results highlight limitations in widely used first-principles methods and call for a reassessment of their predictive power in aqueous systems.

physics.chem-ph↗

arXiv2025

AInsteinBench: Benchmarking Coding Agents on Scientific Repositories

We introduce AInsteinBench, a large-scale benchmark for evaluating whether large language model (LLM) agents can operate as scientific computing development agents within real research software ecosystems. Unlike existing scientific reasoning benchmarks which focus on conceptual knowledge, or software engineering benchmarks that emphasize generic feature implementation and issue resolving, AInsteinBench evaluates models in end-to-end scientific development settings grounded in production-grade scientific repositories. The benchmark consists of tasks derived from maintainer-authored pull requests across six widely used scientific codebases, spanning quantum chemistry, quantum computing, molecular dynamics, numerical relativity, fluid dynamics, and cheminformatics. All benchmark tasks are carefully curated through multi-stage filtering and expert review to ensure scientific challenge, adequate test coverage, and well-calibrated difficulty. By leveraging evaluation in executable environments, scientifically meaningful failure modes, and test-driven verification, AInsteinBench measures a model's ability to move beyond surface-level code generation toward the core competencies required for computational scientific research.

cs.SE↗

arXiv2025

Ab Initio bulk free energy surface of proper ferroelectrics

We report a systematic and accurate approach for deriving the bulk free energy surface (FES), a function of temperature, polarization, and strain, from the first-principles density functional theory (DFT) of proper ferroelectrics. The core of our approach is the metadynamics algorithm that extracts the polarization dependence of the FES from all-atom molecular dynamics simulations without an a priori ansatz. The rest of the FES is derived from the metadynamics trajectories that span the relevant phase space. We demonstrate our approach in the case of lead titanate. The errors across the phase transition, due to DFT numerics, all-atom molecular dynamics, and free energy evaluation by enhanced sampling, can be systematically controlled and are of the order of 1meV/atom. The accuracy of the resulting ab initio FES is only limited by the adopted functional approximation of DFT.

cond-mat.mtrl-sci↗

arXiv2025

TagSplat: Topology-Aware Gaussian Splatting for Dynamic Mesh Modeling and Tracking

Topology-consistent dynamic model sequences are essential for applications such as animation and model editing. However, existing 4D reconstruction methods face challenges in generating high-quality topology-consistent meshes. To address this, we propose a topology-aware dynamic reconstruction framework based on Gaussian Splatting. We introduce a Gaussian topological structure that explicitly encodes spatial connectivity. This structure enables topology-aware densification and pruning, preserving the manifold consistency of the Gaussian representation. Temporal regularization terms further ensure topological coherence over time, while differentiable mesh rasterization improves mesh quality. Experimental results demonstrate that our method reconstructs topology-consistent mesh sequences with significantly higher accuracy than existing approaches. Moreover, the resulting meshes enable precise 3D keypoint tracking. Project page: https://haza628.github.io/tagSplat/

cs.GR↗

arXiv2025

Spin-Adapted Neural Network Wavefunctions in Real Space

Spin plays a fundamental role in understanding electronic structure, yet many real-space wavefunction methods fail to adequately consider it. We introduce the Spin-Adapted Antisymmetrization Method (SAAM), a general procedure that enforces exact total spin symmetry for antisymmetric many-electron wavefunctions in real space. In the context of neural network-based quantum Monte Carlo (NNQMC), SAAM leverages the expressiveness of deep neural networks to capture electron correlation while enforcing exact spin adaptation via group representation theory. This framework provides a principled route to embed physical priors into otherwise black-box neural network wavefunctions, yielding a compact representation of correlated system with neural network orbitals. Compared with existing treatments of spin in NNQMC, SAAM is more accurate and efficient, achieving exact spin purity without any additional tunable hyperparameters. To demonstrate its effectiveness, we apply SAAM to study the spin ladder of iron-sulfur clusters, a long-standing challenge for many-body methods due to their dense spectrum of nearly degenerate spin states. Our results reveal accurate resolution of low-lying spin states and spin gaps in [Fe$_2$S$_2$] and [Fe$_4$S$_4$] clusters, offering new insights into their electronic structures. In sum, these findings establish SAAM as a robust, hyperparameter-free standard for spin-adapted NNQMC, particularly for strongly correlated systems.

physics.chem-ph↗

arXiv2025

Dual-Flow: Transferable Multi-Target, Instance-Agnostic Attacks via In-the-wild Cascading Flow Optimization

Adversarial attacks are widely used to evaluate model robustness, and in black-box scenarios, the transferability of these attacks becomes crucial. Existing generator-based attacks have excellent generalization and transferability due to their instance-agnostic nature. However, when training generators for multi-target tasks, the success rate of transfer attacks is relatively low due to the limitations of the model's capacity. To address these challenges, we propose a novel Dual-Flow framework for multi-target instance-agnostic adversarial attacks, utilizing Cascading Distribution Shift Training to develop an adversarial velocity function. Extensive experiments demonstrate that Dual-Flow significantly improves transferability over previous multi-target generative attacks. For example, it increases the success rate from Inception-v3 to ResNet-152 by 34.58\%. Furthermore, our attack method shows substantially stronger robustness against defense mechanisms, such as adversarially trained models. The code of Dual-Flow is available at: $\href{https://github.com/Chyxx/Dual-Flow}{https://github.com/Chyxx/Dual-Flow}$.

cs.CV↗

arXiv2025

A Trajectory Generator for High-Density Traffic and Diverse Agent-Interaction Scenarios

Accurate trajectory prediction is fundamental to autonomous driving, as it underpins safe motion planning and collision avoidance in complex environments. However, existing benchmark datasets suffer from a pronounced long-tail distribution problem, with most samples drawn from low-density scenarios and simple straight-driving behaviors. This underrepresentation of high-density scenarios and safety critical maneuvers such as lane changes, overtaking and turning is an obstacle to model generalization and leads to overly optimistic evaluations. To address these challenges, we propose a novel trajectory generation framework that simultaneously enhances scenarios density and enriches behavioral diversity. Specifically, our approach converts continuous road environments into a structured grid representation that supports fine-grained path planning, explicit conflict detection, and multi-agent coordination. Built upon this representation, we introduce behavior-aware generation mechanisms that combine rule-based decision triggers with Frenet-based trajectory smoothing and dynamic feasibility constraints. This design allows us to synthesize realistic high-density scenarios and rare behaviors with complex interactions that are often missing in real data. Extensive experiments on the large-scale Argoverse 1 and Argoverse 2 datasets demonstrate that our method significantly improves both agent density and behavior diversity, while preserving motion realism and scenario-level safety. Our synthetic data also benefits downstream trajectory prediction models and enhances performance in challenging high-density scenarios.

cs.RO↗

arXiv2025

Collaborative Compression for Large-Scale MoE Deployment on Edge

The Mixture of Experts (MoE) architecture is an important method for scaling Large Language Models (LLMs). It increases model capacity while keeping computation cost low. However, the ultra-large MoE models still have hundreds of billions of parameters, requiring massive memory/storage and leading to difficulties for deployment on resource-constrained edge platforms. Pruning or quantization alone can hardly address the issue, because of the super-aggressive compression ratio with significantly degraded accuracy and output quality. To facilitate the deployment of ultra-large MoEs on edge platforms, we propose a collaborative compression framework by combining expert pruning, mixed-precision quantization, and activation optimization. It can effectively reduce the storage footprint of the ultra-large MoE DeepSeek-V3 from 1.3TB to 103GB, while preserving high output quality with better accuracy than traditional uniform low-bit quantization methods. To the best of our knowledge, we are the first to deploy a compressed model from the ultra-large DeepSeek-V3 on the platform with a strict 128GB total memory limit. Our comprehensive experiments on multiple benchmarks under various memory constraints demonstrate the effectiveness of our method with smaller model sizes and higher accuracy than uniform low-bit quantization methods.

cs.AI↗

arXiv2025

From Static to Dynamic: a Survey of Topology-Aware Perception in Autonomous Driving

The key to achieving autonomous driving lies in topology-aware perception, the structured understanding of the driving environment with an emphasis on lane topology and road semantics. This survey systematically reviews four core research directions under this theme: vectorized map construction, topological structure modeling, prior knowledge fusion, and language model-based perception. Across these directions, we observe a unifying trend: a paradigm shift from static, pre-built maps to dynamic, sensor-driven perception. Specifically, traditional static maps have provided semantic context for autonomous systems. However, they are costly to construct, difficult to update in real time, and lack generalization across regions, limiting their scalability. In contrast, dynamic representations leverage on-board sensor data for real-time map construction and topology reasoning. Each of the four research directions contributes to this shift through compact spatial modeling, semantic relational reasoning, robust domain knowledge integration, and multimodal scene understanding powered by pre-trained language models. Together, they pave the way for more adaptive, scalable, and explainable autonomous driving systems.

cs.CV↗

arXiv2025

CITER: Collaborative Inference for Efficient Large Language Model Decoding with Token-Level Routing

Large language models have achieved remarkable success in various tasks but suffer from high computational costs during inference, limiting their deployment in resource-constrained applications. To address this issue, we propose a novel Collaborative Inference with Token-lEvel Routing (CITER) framework that enables efficient collaboration between small and large language models (SLMs \& LLMs) through a token-level routing strategy. Specifically, CITER routes non-critical tokens to an SLM for efficiency and routes critical tokens to an LLM for generalization quality. We formulate router training as a policy optimization, where the router receives rewards based on both the quality of predictions and the inference costs of generation. This allows the router to learn to predict token-level routing scores and make routing decisions based on both the current token and the future impact of its decisions. To further accelerate the reward evaluation process, we introduce a shortcut which significantly reduces the costs of the reward estimation and improving the practicality of our approach. Extensive experiments on five benchmark datasets demonstrate that CITER reduces the inference costs while preserving high-quality generation, offering a promising solution for real-time and resource-constrained applications. Our data and code are available at https://github.com/aiming-lab/CITER.

cs.CL↗

arXiv2025

Verbal Werewolf: Engage Users with Verbalized Agentic Werewolf Game Framework

The growing popularity of social deduction games has created an increasing need for intelligent frameworks where humans can collaborate with AI agents, particularly in post-pandemic contexts with heightened psychological and social pressures. Social deduction games like Werewolf, traditionally played through verbal communication, present an ideal application for Large Language Models (LLMs) given their advanced reasoning and conversational capabilities. Prior studies have shown that LLMs can outperform humans in Werewolf games, but their reliance on external modules introduces latency that left their contribution in academic domain only, and omit such game should be user-facing. We propose \textbf{Verbal Werewolf}, a novel LLM-based Werewolf game system that optimizes two parallel pipelines: gameplay powered by state-of-the-art LLMs and a fine-tuned Text-to-Speech (TTS) module that brings text output to life. Our system operates in near real-time without external decision-making modules, leveraging the enhanced reasoning capabilities of modern LLMs like DeepSeek V3 to create a more engaging and anthropomorphic gaming experience that significantly improves user engagement compared to existing text-only frameworks.

cs.CL↗

arXiv2025

RIFLEx: A Free Lunch for Length Extrapolation in Video Diffusion Transformers

Recent advancements in video generation have enabled models to synthesize high-quality, minute-long videos. However, generating even longer videos with temporal coherence remains a major challenge and existing length extrapolation methods lead to temporal repetition or motion deceleration. In this work, we systematically analyze the role of frequency components in positional embeddings and identify an intrinsic frequency that primarily governs extrapolation behavior. Based on this insight, we propose RIFLEx, a minimal yet effective approach that reduces the intrinsic frequency to suppress repetition while preserving motion consistency, without requiring any additional modifications. RIFLEx offers a true free lunch--achieving high-quality 2x extrapolation on state-of-the-art video diffusion transformers in a completely training-free manner. Moreover, it enhances quality and enables 3x extrapolation by minimal fine-tuning without long videos. Project page and codes: https://riflex-video.github.io/.

cs.CV↗

arXiv2025

Neural Scaling Laws Surpass Chemical Accuracy for the Many-Electron Schrödinger Equation

We demonstrate, for the first time, that neural scaling laws can deliver near-exact solutions to the many-electron Schrödinger equation across a broad range of realistic molecules. This progress is enabled by the Lookahead Variational Algorithm (LAVA), an effective optimization scheme that systematically translates increased model size and computational resources into greatly improved energy accuracy for neural network wavefunctions. Across all tested cases, including benzene, the absolute energy error exhibits a systematic power-law decay with respect to model capacity and computation resources. The resulting energies not only surpass the 1 kcal/mol "chemical-accuracy" threshold but also achieve 1 kJ/mol subchemical accuracy. Beyond energies, the scaled-up neural network also yields better wavefunctions with improved physical symmetries, alongside accurate electron densities, dipole moments, and other important properties. Our approach offers a promising way forward to addressing many long-standing challenges in quantum chemistry. For instance, we improve energetic properties for systems such as the potential energy curve of nitrogen dimer as dissociation is approached and the cyclobutadiene automerization reaction barrier, producing definitive benchmarks, particularly in regimes where experimental data are sparse or highly uncertain. We also shed light on the decades-old puzzle of the cyclic ozone stability with highly accurate calculations for the cyclic-to-open ozone barrier. These results provide near-exact reference calculations with unprecedented accuracy, universal reliability and practical applicability, establishing a foundation for AI-driven quantum chemistry.

physics.chem-ph↗

arXiv2025

PS-GS: Gaussian Splatting for Multi-View Photometric Stereo

Integrating inverse rendering with multi-view photometric stereo (MVPS) yields more accurate 3D reconstructions than the inverse rendering approaches that rely on fixed environment illumination. However, efficient inverse rendering with MVPS remains challenging. To fill this gap, we introduce the Gaussian Splatting for Multi-view Photometric Stereo (PS-GS), which efficiently and jointly estimates the geometry, materials, and lighting of the object that is illuminated by diverse directional lights (multi-light). Our method first reconstructs a standard 2D Gaussian splatting model as the initial geometry. Based on the initialization model, it then proceeds with the deferred inverse rendering by the full rendering equation containing a lighting-computing multi-layer perceptron. During the whole optimization, we regularize the rendered normal maps by the uncalibrated photometric stereo estimated normals. We also propose the 2D Gaussian ray-tracing for single directional light to refine the incident lighting. The regularizations and the use of multi-view and multi-light images mitigate the ill-posed problem of inverse rendering. After optimization, the reconstructed object can be used for novel-view synthesis, relighting, and material and shape editing. Experiments on both synthetic and real datasets demonstrate that our method outperforms prior works in terms of reconstruction accuracy and computational efficiency.

cs.GR↗

arXiv2025

Deep variational free energy prediction of dense hydrogen solid at 1200K

We perform deep variational free energy calculations to investigate the dense hydrogen system at 1200 K and high pressures. In this computational framework, neural networks are used to model the free energy through the proton Boltzmann distribution and the electron wavefunction. By directly minimizing the free energy, our results reveal the emergence of a crystalline order associated with the center of mass of hydrogen molecules at approximately 180 GPa. This transition from atomic liquid to a molecular solid is marked by discontinuities in both the pressure and thermal entropy. Additionally, we discuss the broader implications and limitations of these findings in the context of recent studies of dense hydrogen under similar conditions.

physics.comp-ph↗