arXiv subjects

Siyu Li

Publications and source records attributed to Siyu Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

CoBEVMoE: Heterogeneity-aware Feature Fusion with Dynamic Mixture-of-Experts for Collaborative Perception

Collaborative perception aims to extend sensing coverage and improve perception accuracy by sharing information among multiple agents. However, due to differences in viewpoints and spatial positions, agents often acquire heterogeneous observations. Existing intermediate fusion methods primarily focus on aligning similar features, often overlooking the perceptual diversity among agents. To address this limitation, we propose CoBEVMoE, a novel collaborative perception framework that operates in the Bird's Eye View (BEV) space and incorporates a Dynamic Mixture-of-Experts (DMoE) architecture. In DMoE, each expert is dynamically generated based on the input features of a specific agent, enabling it to extract distinctive and reliable cues while attending to shared semantics. This design allows the fusion process to explicitly model both feature similarity and heterogeneity across agents. Furthermore, we introduce a Dynamic Expert Metric Loss (DEML) to enhance inter-expert diversity and improve the discriminability of the fused representation. Extensive experiments on the OPV2V and DAIR-V2X-C datasets demonstrate that CoBEVMoE achieves state-of-the-art performance. Specifically, it improves the IoU for Camera-based BEV segmentation by +1.5% on OPV2V and the AP@0.5 for LiDAR-based 3D object detection by +3.0% on DAIR-V2X-C, verifying the effectiveness of expert-based heterogeneous feature modeling in multi-agent collaborative perception. The source code will be made publicly available at https://github.com/godk0509/CoBEVMoE.

cs.CV↗

arXiv2026

Graph-Augmented Reasoning with Large Language Models for Tobacco Pest and Disease Management

This paper proposes a graph-augmented reasoning framework for tobacco pest and disease management that integrates structured domain knowledge into large language models. Building on GraphRAG, we construct a domain-specific knowledge graph and retrieve query-relevant subgraphs to provide relational evidence during answer generation. The framework adopts ChatGLM as the Transformer backbone with LoRA-based parameter-efficient fine-tuning, and employs a graph neural network to learn node representations that capture symptom-disease-treatment dependencies. By explicitly modeling diseases, symptoms, pesticides, and control measures as linked entities, the system supports evidence-aware retrieval beyond surface-level text similarity. Retrieved graph evidence is incorporated into the LLM input to guide generation toward domain-consistent recommendations and to mitigate hallucinated or inappropriate treatments. Experimental results show consistent improvements over text-only baselines, with the largest gains observed on multi-hop and comparative reasoning questions that require chaining multiple relations.

cs.CL↗

arXiv2026

Inference-Time Scaling for Visual AutoRegressive modeling by Searching Representative Samples

While inference-time scaling has significantly enhanced generative quality in large language and diffusion models, its application to vector-quantized (VQ) visual autoregressive modeling (VAR) remains unexplored. We introduce VAR-Scaling, the first general framework for inference-time scaling in VAR, addressing the critical challenge of discrete latent spaces that prohibit continuous path search. We find that VAR scales exhibit two distinct pattern types: general patterns and specific patterns, where later-stage specific patterns conditionally optimize early-stage general patterns. To overcome the discrete latent space barrier in VQ models, we map sampling spaces to quasi-continuous feature spaces via kernel density estimation (KDE), where high-density samples approximate stable, high-quality solutions. This transformation enables effective navigation of sampling distributions. We propose a density-adaptive hybrid sampling strategy: Top-k sampling focuses on high-density regions to preserve quality near distribution modes, while Random-k sampling explores low-density areas to maintain diversity and prevent premature convergence. Consequently, VAR-Scaling optimizes sample fidelity at critical scales to enhance output quality. Experiments in class-conditional and text-to-image evaluations demonstrate significant improvements in inference process. The code is available at https://github.com/WD7ang/VAR-Scaling.

cs.CV↗

arXiv2025

Knowledge Reasoning of Large Language Models Integrating Graph-Structured Information for Pest and Disease Control in Tobacco

This paper proposes a large language model (LLM) approach that integrates graph-structured information for knowledge reasoning in tobacco pest and disease control. Built upon the GraphRAG framework, the proposed method enhances knowledge retrieval and reasoning by explicitly incorporating structured information from a domain-specific knowledge graph. Specifically, LLMs are first leveraged to assist in the construction of a tobacco pest and disease knowledge graph, which organizes key entities such as diseases, symptoms, control methods, and their relationships. Based on this graph, relevant knowledge is retrieved and integrated into the reasoning process to support accurate answer generation. The Transformer architecture is adopted as the core inference model, while a graph neural network (GNN) is employed to learn expressive node representations that capture both local and global relational information within the knowledge graph. A ChatGLM-based model serves as the backbone LLM and is fine-tuned using LoRA to achieve parameter-efficient adaptation. Extensive experimental results demonstrate that the proposed approach consistently outperforms baseline methods across multiple evaluation metrics, significantly improving both the accuracy and depth of reasoning, particularly in complex multi-hop and comparative reasoning scenarios.

cs.CL↗

arXiv2025

A perceptual bias of AI Logical Argumentation Ability in Writing

Can machines think? This is a central question in artificial intelligence research. However, there is a substantial divergence of views on the answer to this question. Why do people have such significant differences of opinion, even when they are observing the same real world performance of artificial intelligence? The ability of logical reasoning like humans is often used as a criterion to assess whether a machine can think. This study explores whether human biases influence evaluations of the reasoning abilities of AI. An experiment was conducted where participants assessed two texts on the same topic, one AI generated and one human written,to test for perceptual biases in evaluating logical reasoning. Based on the experimental findings, a questionnaire was designed to quantify the attitudes toward AI.The results reveal a bias in perception. The evaluations of the logical reasoning ability of AI generated texts are significantly influenced by the preconceived views on the logical reasoning abilities of AI. Furthermore, frequent AI users were less likely to believe that AI usage undermines independent thinking.This study highlights the need to address perceptual biases to improve public understanding of AI's capabilities and foster better human AI interactions.

cs.AI↗

arXiv2025

New quasi-exactly solvable systems from SUSYQM and Bethe Ansatz

We give a systematic construction of new quasi-exactly solvable systems via Bethe ansatz and supersymmetric quantum mechanics (SUSYQM). Methods based on the intertwining of supercharges have been extensively used in the literature for exactly solvable systems. We generalize the state-deleting (Krein-Adler) supersymmetric transformations to quasi-exactly exactly solvable (QES) systems building on the Bethe ansatz method and related Bethe roots. This enables us to construct superpartners for a wide class of known QES systems classified previously through a hidden $sl(2)$ algebra. We present our constructions of factorizations and intertwining relations related to 1st-order SUSYQM and the $n=1$ state for 10 nonequivalent types, denoted I,...,X. In order to have a unified treatment we rely on their ODE standard form as this is also the appropriate setting to obtain the Bethe ansatz equations which constrain the polynomial solutions. This setting also allows one to deal with systems with $n$ states in a unified manner, using analysis based on the Bethe ansatz equations to build the supersymmetric transformations in terms of the Bethe ansatz roots. We derive the Schrödinger potentials for the $n=1$ superpartners of the 10 QES cases and give closed-form solutions for the spectra and wavefunctions of the corresponding QES SUSYQM systems. Furthermore, we present numerical results for higher excited states up to the $n=10$ level. The results obtained may have wider applicability as our framework is built on ODEs with polynomial coefficients.

math-ph↗

arXiv2025

Enhancing Primordial B-mode Detection: Comprehensive Delensing Pipelines for Improved Sensitivity to $r$

Recognizing the impact of contamination from weak gravitational lensing B-modes induced by Large Scale Structure, we examine delensing methods to enhance sensitivity to the tensor-to-scalar ratio $r$ in primordial B-mode detection experiments. This study presents a realistic pipeline to improve $r$ constraints using foreground-cleaned maps with negligible residuals. The pipeline, based on simulations, is adaptable for future experiments. We focus on two delensing approaches: (1) subtracting the gradient-order lensing B-mode template, computed by convolving the E-mode with the lensing potential, from the observed B-mode signal; and (2) remapping observations using the estimated inverse deflection angle. For parameter constraints, we employ three models to reduce $r$ uncertainty and bias, finding consistent uncertainties across models, though biases vary due to the multipole-dependence of the delensing fraction. %We demonstrate the pipeline using simulated maps from future CMB polarization experiments, including a small-aperture (sub-1m) telescope, a large-aperture (6m) telescope, and a future space mission (3m). We demonstrated this pipeline using simulated observation maps from future CMB polarization experiments, which included current representative ground-based small aperture telescopes (sub-1m), next-generation ground-based large aperture telescopes (6m), and highly competitive future space-based medium aperture missions (3m). Results show a delensing efficiency of 40\% with the small-aperture telescope alone, increasing to 65\% when combined with the large-aperture telescope, and 80\% with the satellite mission. These lead to reductions in $r$ uncertainty by 46\% for ground-based and 63\% for space missions. The most promising method adds the lensing template B-mode as an additional frequency channel, minimizing bias on $r$.

astro-ph.CO↗

arXiv2025

Virtual Width Networks

We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN decouples representational width from backbone width, expanding the embedding space while keeping backbone compute nearly constant. In our large-scale experiment, an 8-times expansion accelerates optimization by over 2 times for next-token and 3 times for next-2-token prediction. The advantage amplifies over training as both the loss gap grows and the convergence-speedup ratio increases, showing that VWN is not only token-efficient but also increasingly effective with scale. Moreover, we identify an approximately log-linear scaling relation between virtual width and loss reduction, offering an initial empirical basis and motivation for exploring virtual-width scaling as a new dimension of large-model efficiency.

cs.LG↗

arXiv2025

MIO: A Foundation Model on Multimodal Tokens

In this paper, we introduce MIO, a novel foundation model built on multimodal tokens, capable of understanding and generating speech, text, images, and videos in an end-to-end, autoregressive manner. While the emergence of large language models (LLMs) and multimodal large language models (MM-LLMs) propels advancements in artificial general intelligence through their versatile capabilities, they still lack true any-to-any understanding and generation. Recently, the release of GPT-4o has showcased the remarkable potential of any-to-any LLMs for complex real-world tasks, enabling omnidirectional input and output across images, speech, and text. However, it is closed-source and does not support the generation of multimodal interleaved sequences. To address this gap, we present MIO, which is trained on a mixture of discrete tokens across four modalities using causal multimodal modeling. MIO undergoes a four-stage training process: (1) alignment pre-training, (2) interleaved pre-training, (3) speech-enhanced pre-training, and (4) comprehensive supervised fine-tuning on diverse textual, visual, and speech tasks. Our experimental results indicate that MIO exhibits competitive, and in some cases superior, performance compared to previous dual-modal baselines, any-to-any model baselines, and even modality-specific baselines. Moreover, MIO demonstrates advanced capabilities inherent to its any-to-any feature, such as interleaved video-text generation, chain-of-visual-thought reasoning, visual guideline generation, instructional image editing, etc.

cs.CL↗

arXiv2025

Polynomial deformations of $sl(2)$ and unified algebraic framework for solutions of a class of spin models

We introduce novel polynomial deformations of the Lie algebra $sl(2)$. We construct their finite-dimensional irreducible representations and the corresponding differential operator realizations. We apply our results to a class of spin models with hidden polynomial algebra symmetry and obtain the closed-form expressions for their energies and wave functions by means of the Bethe ansatz method. The general framework enables us to give an unified algebraic and analytic treatment for three interesting spin models with hidden cubic algebra symmetry: the Lipkin-Meshkov-Glick (LMG) model, the molecular asymmetric rigid rotor, and the two-axis countertwisting squeezing model. We provide analytic and numerical insights into the structures of the roots of the Bethe ansatz equations (i.e. the so-called Bethe roots) of these models. We give descriptions of the roots on the spheres using the inverse stereographic projection. The changes in nature and pattern of the Bethe roots on the spheres indicate the existence of different phases of the models. We also present the fidelity and derivatives of the ground-state energies (with respect to model parameters) of the models. The results indicate the presence of critical points and phase transitions of the models. In the appendix, we show that, unlike the so-called Bender-Dunne polynomials, the set of polynomials in the energy $E$, $P_\ell(E)$, corresponding to each of the three spin models has two critical polynomials whose zeros give the quasi-exact energy eigenvalues of the model. Such types of polynomials seem new.

math-ph↗

arXiv2025

Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learning

We introduce Seed1.5-Thinking, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed1.5-Thinking achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research. Model trial link: https://www.volcengine.com/experience/ark.

cs.CL↗

arXiv2025

Geo-Llama: Leveraging LLMs for Human Mobility Trajectory Generation with Spatiotemporal Constraints

Generating realistic human mobility data is essential for various application domains, including transportation, urban planning, and epidemic control, as real data is often inaccessible to researchers due to high costs and privacy concerns. Existing deep generative models learn from real trajectories to generate synthetic ones. Despite the progress, most of them suffer from training stability issues and scale poorly with increasing data size. More importantly, they often lack control mechanisms to guide the generated trajectories under constraints such as enforcing specific visits. To address these limitations, we formally define the controlled trajectory generation problem for effectively handling multiple spatiotemporal constraints. We introduce Geo-Llama, a novel LLM finetuning framework that can enforce multiple explicit visit constraints while maintaining contextual coherence of the generated trajectories. In this approach, pre-trained LLMs are fine-tuned on trajectory data with a visit-wise permutation strategy where each visit corresponds to a specific time and location. This strategy enables the model to capture spatiotemporal patterns regardless of visit orders while maintaining flexible and in-context constraint integration through prompts during generation. Extensive experiments on real-world and synthetic datasets validate the effectiveness of Geo-Llama, demonstrating its versatility and robustness in handling a broad range of constraints to generate more realistic trajectories compared to existing methods.

cs.AI↗

arXiv2025

Stäckel transform, coupling constant metamorphosis and algebraization of quasi-exactly solvable systems

We generalize the notions of the Stäckel transform and the coupling constant metamorphosis to quasi-exactly solvable systems. We discover that for a variety of one-dimensional and separable multidimensional quasi-exactly solvable systems, their $sl(2)$ algebraizations can only be achieved via coupling constant metamorphosis after appropriate Stäckel transformations. This discovery has interesting applications, allowing us to derive algebraizations and energies for a wide class of quasi-exactly solvable systems, such as Hooke's atoms in magnetic fields and Newtonian cosmology. The approach of coupling constant metamorphosis was successfully applied previously in the context of exactly solvable, integrable and superintegrable systems. To our knowledge, the present work is the first to apply the idea and approach in the context of quasi-exactly solvable systems.

math-ph↗

arXiv2025

Impact of the RNA Binding Domain Localization of the Protein Shell on Virus Particle Stability

The encapsulation of polyanions, whether single-stranded RNAs or synthetic polymers, is primarily driven by attractive electrostatic interactions between the positively charged, structurally disordered RNA-binding domains of virus coat proteins and the negatively charged polyanions. Theoretically, this interaction is often modeled by coarse-graining the charge distribution of the binding domains, either by projecting the charges onto the inner surface of the protein shell or by spreading them across a region representing the capsid lumen where the binding domains are located. In practice, however, the positive charges are not uniformly distributed across the binding domains, which themselves are positioned at discrete, specific sites on the shell surface. Here, we use molecular dynamics simulations to investigate the impact of localized interactions on the most probable or optimal length of the encapsulated polymer, revealing that the specific location of charges along the binding domains plays a significant role, consistent with experimental observations. Comparing the simulations with predictions from a simple mean-field theory taken from the literature, we find that while the general trends are reasonably well captured, quantitative discrepancies arise between the two approaches.

physics.bio-ph↗

arXiv2025

General Solution for Elastic Networks on Arbitrary Curved Surfaces in the Absence of Rotational Symmetry

Understanding crystal growth over arbitrary curved surfaces with arbitrary boundaries is a formidable challenge, stemming from the complexity of formulating non-linear elasticity using geometric invariant quantities. Solutions are generally confined to systems exhibiting rotational symmetry. In this paper, we introduce a framework to address these challenges by numerically solving these equations without relying on inherent symmetries. We illustrate our approach by computing the minimum energy required for an elastic network containing a disclination at any point and by investigating surfaces that lack rotational symmetry. Our findings reveal that the transition from a defect-free structure to a stable state with a single 5-fold or 7-fold disclination strongly depends on the shape of the domain, emphasizing the profound influence of edge geometry. We discuss the implications of our results for general experimental systems, particularly in elucidating the assembly pathways of virus capsids. This research enhances our understanding of crystal growth on complex surfaces and expands its applications across diverse scientific domains.

cond-mat.soft↗

arXiv2024

Proposing and solving olympiad geometry with guided tree search

Mathematics olympiads are prestigious competitions, with problem proposing and solving highly honored. Building artificial intelligence that proposes and solves olympiads presents an unresolved challenge in automated theorem discovery and proving, especially in geometry for its combination of numerical and spatial elements. We introduce TongGeometry, a Euclidean geometry system supporting tree-search-based guided problem proposing and solving. The efficient geometry system establishes the most extensive repository of geometry theorems to date: within the same computational budget as the existing state-of-the-art, TongGeometry discovers 6.7 billion geometry theorems requiring auxiliary constructions, including 4.1 billion exhibiting geometric symmetry. Among them, 10 theorems were proposed to regional mathematical olympiads with 3 of TongGeometry's proposals selected in real competitions, earning spots in a national team qualifying exam or a top civil olympiad in China and the US. Guided by fine-tuned large language models, TongGeometry solved all International Mathematical Olympiad geometry in IMO-AG-30, outperforming gold medalists for the first time. It also surpasses the existing state-of-the-art across a broader spectrum of olympiad-level problems. The full capabilities of the system can be utilized on a consumer-grade machine, making the model more accessible and fostering widespread democratization of its use. By analogy, unlike existing systems that merely solve problems like students, TongGeometry acts like a geometry coach, discovering, presenting, and proving theorems.

cs.AI↗

arXiv2024

Group-Control Motion Planning Framework for Microrobot Swarms in a Global Field

This paper investigates how a novel paradigm called group-control can be effectively used for motion planning for microrobot swarms in a global field. We prove that Small-Time Local Controllability (STLC) in robot positions is achievable through group-control, with the minimum number of groups required for STLC being $\log_2(n + 2) + 1$ for $n$ robots. We then discuss the trade-offs between control complexity, motion planning complexity, and motion efficiency. We show how motion planning can be simplified if appropriate primitives can be achieved through more complex control actions. We identify motion planning problems that balance the number of motion primitives with planning complexity. Various instantiations of these motion planning problems are explored, with simulations used to demonstrate the effectiveness of group-control.

cs.RO↗

arXiv2024

Trans4D: Realistic Geometry-Aware Transition for Compositional Text-to-4D Synthesis

Recent advances in diffusion models have demonstrated exceptional capabilities in image and video generation, further improving the effectiveness of 4D synthesis. Existing 4D generation methods can generate high-quality 4D objects or scenes based on user-friendly conditions, benefiting the gaming and video industries. However, these methods struggle to synthesize significant object deformation of complex 4D transitions and interactions within scenes. To address this challenge, we propose Trans4D, a novel text-to-4D synthesis framework that enables realistic complex scene transitions. Specifically, we first use multi-modal large language models (MLLMs) to produce a physic-aware scene description for 4D scene initialization and effective transition timing planning. Then we propose a geometry-aware 4D transition network to realize a complex scene-level 4D transition based on the plan, which involves expressive geometrical object deformation. Extensive experiments demonstrate that Trans4D consistently outperforms existing state-of-the-art methods in generating 4D scenes with accurate and high-quality transitions, validating its effectiveness. Code: https://github.com/YangLing0818/Trans4D

cs.CV↗