arXiv subjects

Shiyu Fan

Publications and source records attributed to Shiyu Fan.

17 recordsLinked to original sources

arXiv2026

Generative Motion In-betweening by Diffusion over Continuous Implicit Representations

Recent advances in generative models have yielded impressive progress on motion in-betweening, allowing for more complex, varied, and realistic motion transitions. However, recent methods still exhibit noticeable limitations in preserving keyframe information and ensuring motion continuity. In this paper, we propose a novel pipeline and sampling optimization strategy for latent diffusion models (LDM) based on motion implicit neural representations (INR). By establishing a mapping between INR and sparse spatial or temporal information within latent diffusion, our model can sample the INR parameters from extremely sparse and ambiguous keyframe data and reconstruct plausible and smooth motions from the manifold. Our experiments demonstrate the superior performance of our model, which significantly improves motion generation quality in scenarios with few keyframes while ensuring both keyframe accuracy and diversity of in-between motions.

cs.GR↗

arXiv2026

Excitations across the equilibrium and photoinduced `hidden' states of magnetoresistive manganites

"Hidden" phases, generated using ultrafast laser pulses (few hundred femtoseconds), with properties distinct from thermodynamic equilibrium, are appealing for technologies because they can be long-lived, with lifetimes of hours or weeks, and reversible with temperature sweeping or extra pulses. In this regard, La$_{2/3}$Ca$_{1/3}$MnO$_3$ (LCMO) stands out due to its tunability through epitaxial strain, which can drive the bulk ferromagnetic metal (FMM) into an antiferromagnetic insulator (AFI), and its susceptibility to photo-induced transitions. Indeed, AFI LCMO displays a long-lived photo-induced transition into a putative 'hidden' phase whose exact nature and excitations are still largely unknown. Here, we combine ultrafast photo-excitation in the near infrared with in situ transport, x-ray absorption (XAS), and Resonant Inelastic X-ray Scattering (RIXS) to investigate the excitations (polarons, phonons, and orbital) of the photo-excited phase of LCMO and contrast them with the thermodynamic phases achieved through strain and temperature. In the thermodynamic regime, we establish the correlation between polarons and transport, placing them in the 'strong coupling' regime of the Holstein model. Upon photo-excitation of LCMO-AFI, we uncover a long-lived phase characterized by the softening of the polaron excitations, the partial suppression of the Jahn-Teller distortion, and nearly unchanged phonons, showing the emergence of a photo-excited state absent in the equilibrium phase diagram. Finally, by varying temperature, epitaxial strain, and photo-excitation fluence, we construct a polaron phase diagram and identify the key spectroscopic signatures of each phase. Our laser-RIXS approach establishes a versatile platform for exploring photo-induced 'hidden' phases in quantum materials in non-stroboscopic conditions.

cond-mat.str-el↗

arXiv2025

Masked Generative Policy for Robotic Control

We present Masked Generative Policy (MGP), a novel framework for visuomotor imitation learning. We represent actions as discrete tokens, and train a conditional masked transformer that generates tokens in parallel and then rapidly refines only low-confidence tokens. We further propose two new sampling paradigms: MGP-Short, which performs parallel masked generation with score-based refinement for Markovian tasks, and MGP-Long, which predicts full trajectories in a single pass and dynamically refines low-confidence action tokens based on new observations. With globally coherent prediction and robust adaptive execution capabilities, MGP-Long enables reliable control on complex and non-Markovian tasks that prior methods struggle with. Extensive evaluations on 150 robotic manipulation tasks spanning the Meta-World and LIBERO benchmarks show that MGP achieves both rapid inference and superior success rates compared to state-of-the-art diffusion and autoregressive policies. Specifically, MGP increases the average success rate by 9% across 150 tasks while cutting per-sequence inference time by up to 35x. It further improves the average success rate by 60% in dynamic and missing-observation environments, and solves two non-Markovian scenarios where other state-of-the-art methods fail.

cs.RO↗

arXiv2025

Structural origin of resonant diffraction in RuO$_2$

We report Ru L$_3$-edge resonant X-ray diffraction studies on single crystal and (001) epitaxial films of RuO$_2$. We investigate the distinct $\mathbf{Q} = (100)$ and $(001)$ reflections as a function of incident energy, azimuthal angle, and temperature. The results show that the observed resonant diffraction in RuO$_2$ is fully consistent with a resonant charge anisotropy signal of structural origin permitted by the parent (non-magnetic) rutile $P4_2/mnm$ space group. These results significantly constrain the magnetic contribution to the resonant diffraction signal and indicate the unlikely existence of $\mathbf{k} = 0$ antiferromagnetic order in RuO$_2$.

cond-mat.str-el↗

arXiv2025

Leveraging LLMs for Automated Translation of Legacy Code: A Case Study on PL/SQL to Java Transformation

The VT legacy system, comprising approximately 2.5 million lines of PL/SQL code, lacks consistent documentation and automated tests, posing significant challenges for refactoring and modernisation. This study investigates the feasibility of leveraging large language models (LLMs) to assist in translating PL/SQL code into Java for the modernised "VTF3" system. By leveraging a dataset comprising 10 PL/SQL-to-Java code pairs and 15 Java classes, which collectively established a domain model for the translated files, multiple LLMs were evaluated. Furthermore, we propose a customized prompting strategy that integrates chain-of-guidance reasoning with $n$-shot prompting. Our findings indicate that this methodology effectively guides LLMs in generating syntactically accurate translations while also achieving functional correctness. However, the findings are limited by the small sample size of available code files and the restricted access to test cases used for validating the correctness of the generated code. Nevertheless, these findings lay the groundwork for scalable, automated solutions in modernising large legacy systems.

cs.SE↗

arXiv2025

Waymo-3DSkelMo: A Multi-Agent 3D Skeletal Motion Dataset for Pedestrian Interaction Modeling in Autonomous Driving

Large-scale high-quality 3D motion datasets with multi-person interactions are crucial for data-driven models in autonomous driving to achieve fine-grained pedestrian interaction understanding in dynamic urban environments. However, existing datasets mostly rely on estimating 3D poses from monocular RGB video frames, which suffer from occlusion and lack of temporal continuity, thus resulting in unrealistic and low-quality human motion. In this paper, we introduce Waymo-3DSkelMo, the first large-scale dataset providing high-quality, temporally coherent 3D skeletal motions with explicit interaction semantics, derived from the Waymo Perception dataset. Our key insight is to utilize 3D human body shape and motion priors to enhance the quality of the 3D pose sequences extracted from the raw LiDRA point clouds. The dataset covers over 14,000 seconds across more than 800 real driving scenarios, including rich interactions among an average of 27 agents per scene (with up to 250 agents in the largest scene). Furthermore, we establish 3D pose forecasting benchmarks under varying pedestrian densities, and the results demonstrate its value as a foundational resource for future research on fine-grained human behavior understanding in complex urban environments. The dataset and code will be available at https://github.com/GuangxunZhu/Waymo-3DSkelMo

cs.CV↗

arXiv2025

Multi-Person Interaction Generation from Two-Person Motion Priors

Generating realistic human motion with high-level controls is a crucial task for social understanding, robotics, and animation. With high-quality MOCAP data becoming more available recently, a wide range of data-driven approaches have been presented. However, modelling multi-person interactions still remains a less explored area. In this paper, we present Graph-driven Interaction Sampling, a method that can generate realistic and diverse multi-person interactions by leveraging existing two-person motion diffusion models as motion priors. Instead of training a new model specific to multi-person interaction synthesis, our key insight is to spatially and temporally separate complex multi-person interactions into a graph structure of two-person interactions, which we name the Pairwise Interaction Graph. We thus decompose the generation task into simultaneous single-person motion generation conditioned on one other's motion. In addition, to reduce artifacts such as interpenetrations of body parts in generated multi-person interactions, we introduce two graph-dependent guidance terms into the diffusion sampling scheme. Unlike previous work, our method can produce various high-quality multi-person interactions without having repetitive individual motions. Extensive experiments demonstrate that our approach consistently outperforms existing methods in reducing artifacts when generating a wide range of two-person and multi-person interactions.

cs.GR↗

arXiv2025

Topotactic Reduction-Driven Crystal Field Excitations in Brownmillerite Manganite Thin Films

Topotactic reduction of perovskite oxides offers a powerful approach for discovering novel phenomena, such as superconducting infinite-layer nickelates and polar metallicity, and is commonly accompanied by the emergence of multiple valence states and/or complex crystal fields of transition metals. However, understanding the complex interplay between crystal chemistry, electronic structure, and physical properties at the spin- and orbital-resolved levels in these reduced systems remains elusive. Here, we combine x-ray absorption spectroscopy, resonant inelastic x-ray scattering (RIXS), and density functional theory calculations to uncover topotactic metal-insulator transition and orbital-specific crystal field excitations in brownmillerite La0.67Ca0.33MnO2.5 thin films. We reveal the Mn valence states to be predominantly Mn2+/Mn3+, along with their corresponding populations at octahedral and tetrahedral sites, which effectively weaken the Mn-O hybridization compared to the parent perovskite phase. As a result, La0.67Ca0.33MnO2.5 films exhibit an antiferromagnetic insulating ground state. Moreover, by combining the RIXS measurements on selected single-valence manganites, specifically MnO, LaMnO3, and CaMnO3, with orbital- and spin-resolved density-of-states calculations, we identify the dd excitations of octahedrally and tetrahedrally coordinated Mn2+/Mn3+ ions, directly linking the microscopic electronic structure to the macroscopic magnetic/electrical properties.

cond-mat.mtrl-sci↗

arXiv2024

Flat'n'Fold: A Diverse Multi-Modal Dataset for Garment Perception and Manipulation

We present Flat'n'Fold, a novel large-scale dataset for garment manipulation that addresses critical gaps in existing datasets. Comprising 1,212 human and 887 robot demonstrations of flattening and folding 44 unique garments across 8 categories, Flat'n'Fold surpasses prior datasets in size, scope, and diversity. Our dataset uniquely captures the entire manipulation process from crumpled to folded states, providing synchronized multi-view RGB-D images, point clouds, and action data, including hand or gripper positions and rotations. We quantify the dataset's diversity and complexity compared to existing benchmarks and show that our dataset features natural and diverse manipulations of real-world demonstrations of human and robot demonstrations in terms of visual and action information. To showcase Flat'n'Fold's utility, we establish new benchmarks for grasping point prediction and subtask decomposition. Our evaluation of state-of-the-art models on these tasks reveals significant room for improvement. This underscores Flat'n'Fold's potential to drive advances in robotic perception and manipulation of deformable objects. Our dataset can be downloaded at https://cvas-ug.github.io/flat-n-fold

cs.RO↗

arXiv2024

Highly Tunable Ru-dimer Molecular Orbital State in 6H-perovskite Ba$_3$MRu$_2$O$_9$

Molecular orbital (MO) systems with clusters of heavy transition metal (TM) ions are one of the most important classes of model materials for studying the interplay between local physics and effects of itinerancy. Despite a large number of candidates identified in the family of 4d TM materials, an understanding of their physics from competing \textit{microscopic} energy scales is still missing. We bridge this gap by reporting the first resonant inelastic X-ray scattering (RIXS) measurement on a well-known series of Ru dimer systems with a 6H-perovskite structure, Ba$_3$MRu$_2$O$_9$ (M$^{3+}$=In$^{3+}$, Y$^{3+}$, La$^{3+}$). Our RIXS measurements reveal an extremely fragile MO state in these Ru dimer compounds, evidenced by an abrupt change in the RIXS spectrum accompanying a tiny change in the local structure tuned by the M-site ion. By modelling the RIXS spectra, we attribute the enhanced electronic instability in Ba$_3$MRu$_2$O$_9$ to the combined effect of a large hopping and a small spin-orbit coupling in the Ru dimers. The unique combination of energy scales uncovered in the present study make Ru MO systems ideal model systems for studying quantum phase transitions with molecular orbitals.

cond-mat.str-el↗

arXiv2024

Elementary excitations of single-photon emitters in hexagonal Boron Nitride

Single-photon emitters serve as building blocks for many emerging concepts in quantum photonics. The recent identification of bright, tunable, and stable emitters in hexagonal boron nitride (hBN) has opened the door to quantum platforms operating across the infrared to ultraviolet spectrum. While it is widely acknowledged that defects are responsible for single-photon emitters in hBN, crucial details regarding their origin, electronic levels, and orbital involvement remain unknown. Here, we employ a combination of resonant inelastic X-ray scattering and photoluminescence spectroscopy in defective hBN unveiling an elementary excitation at 285 meV that gives rise to a plethora of harmonics correlated with single-photon emitters. We discuss the importance of N $\pi^*$ antibonding orbitals in shaping the electronic states of the emitters. The discovery of the elementary excitations of hBN provides new fundamental insights into quantum emission in low-dimensional materials, paving the way for future investigations in other platforms.

cond-mat.mes-hall↗

arXiv2023

Missing-modality Enabled Multi-modal Fusion Architecture for Medical Data

Fusing multi-modal data can improve the performance of deep learning models. However, missing modalities are common for medical data due to patients' specificity, which is detrimental to the performance of multi-modal models in applications. Therefore, it is critical to adapt the models to missing modalities. This study aimed to develop an efficient multi-modal fusion architecture for medical data that was robust to missing modalities and further improved the performance on disease diagnosis.X-ray chest radiographs for the image modality, radiology reports for the text modality, and structured value data for the tabular data modality were fused in this study. Each modality pair was fused with a Transformer-based bi-modal fusion module, and the three bi-modal fusion modules were then combined into a tri-modal fusion framework. Additionally, multivariate loss functions were introduced into the training process to improve model's robustness to missing modalities in the inference process. Finally, we designed comparison and ablation experiments for validating the effectiveness of the fusion, the robustness to missing modalities and the enhancements from each key component. Experiments were conducted on MIMIC-IV, MIMIC-CXR with the 14-label disease diagnosis task. Areas under the receiver operating characteristic curve (AUROC), the area under the precision-recall curve (AUPRC) were used to evaluate models' performance. The experimental results demonstrated that our proposed multi-modal fusion architecture effectively fused three modalities and showed strong robustness to missing modalities. This method is hopeful to be scaled to more modalities to enhance the clinical practicality of the model.

eess.IV↗

arXiv2022

Magnetic Excitations in Strained Infinite-layer Nickelate PrNiO2

Strongly correlated materials often respond sensitively to the external perturbations. In the recently discovered superconducting infinite-layer nickelates, the superconducting transition temperature can be dramatically enhanced via only ~1% compressive strain-tuning enabled by substrate design. However, the root of such enhancement remains elusive. While the superconducting pairing mechanism is still not settled, magnetic Cooper pairing - similar to the cuprates has been proposed. Using resonant inelastic x-ray scattering, we investigate the magnetic excitations in infinite-layer PrNiO2 thin films for different strain conditions. The magnon bandwidth of PrNiO2 shows only marginal response to strain-tuning, in sharp contrast to the striking enhancement of the superconducting transition temperature Tc in the doped superconducting samples. These results suggest the enhancement of Tc is not mediated by spin excitations and thus provide important empirics for the understanding of superconductivity in infinite-layer nickelates.

cond-mat.supr-con↗

arXiv2022

Strain-modulated anisotropic electronic structure in superconducting RuO$_2$ films

The binary ruthenate, RuO$_2$, has been the subject of intense interest due to its itinerant antiferromagnetism and strain-induced superconductivity. The strain mechanism and its effect on the microscopic electronic states leading to the normal and superconducting state, however, remain undisclosed. Here, we investigate highly-strained epitaxial (110) RuO$_2$ films using polarization-dependent oxygen K-edge X-ray absorption spectroscopy (XAS). Through the detection of pre-edge peaks, arising from O:$2p$ - Ru:$4d$ hybridization, we uncover the effects of epitaxial strain on the orbital/electronic structure near the Fermi level. Our data show robust strain-induced shifts of orbital levels and a reduction of hybridization strength. Furthermore, we reveal a pronounced in-plane anisotropy of the electronic structure along the $[110]/[1\bar{1}0]$ directions naturally stemming from the symmetry-breaking epitaxial strain of the substrate. The $B_{2g}$ symmetry component of the epitaxially-enforced strain breaks a sublattice degeneracy, resulting in an increase of the density of states at the Fermi level ($E_F$), possibly paving the way to superconductivity. These results underscore the importance of the effective reduction from tetragonal to orthorhombic lattice symmetry in (110) RuO$_2$ films and its relevance towards the superconducting and magnetic properties.

cond-mat.supr-con↗

arXiv2022

Vibrational fingerprints of ferroelectric hafnia

Hafnia (HfO2) is a promising material for emerging chip applications due to its high-k dielectric behaviour, suitability for negative capacitance heterostructures, scalable ferroelectricity, and silicon compatibility. The lattice dynamics along with phononic properties such as thermal conductivity, contraction, and heat capacity are under-explored, primarily due to the absence of high quality single crystals. Herein, we report the vibrational properties of a series of HfO2 crystals stabilized with yttrium (chemical formula HfO2:xY, where x = 20, 12, 11, 8, and 0%) and compare our findings with a symmetry analysis and lattice dynamics calculations. We untangle the effects of Y by testing our calculations against the measured Raman and infrared spectra of the cubic, antipolar orthorhombic, and monoclinic phases and then proceed to reveal the signature modes of polar orthorhombic hafnia. This work provides a spectroscopic fingerprint for several different phases of HfO2 and paves the way for an analysis of mode contributions to high-k dielectric and ferroelectric properties for chip technologies.

cond-mat.mtrl-sci↗

arXiv2021

A Fast Location Algorithm for Very Sparse Point Clouds Based on Object Detection

Limited by the performance factor, it is arduous to recognize target object and locate it in Augmented Reality (AR) scenes on low-end mobile devices, especially which using monocular cameras. In this paper, we proposed an algorithm which can quickly locate the target object through image object detection in the circumstances of having very sparse feature points. We introduce YOLOv3-Tiny to our algorithm as the object detection module to filter the possible points and using Principal Component Analysis (PCA) to determine the location. We conduct the experiment in a manually designed scene by holding a smartphone and the results represent high positioning speed and accuracy of our method.

cs.CV↗

arXiv2021

A Medical Pre-Diagnosis System for Histopathological Image of Breast Cancer

This paper constructs a novel intelligent medical diagnosis system, which can realize automatic communication and breast cancer pathological image recognition. This system contains two main parts, including a pre-training chatbot called M-Chatbot and an improved neural network model of EfficientNetV2-S named EfficientNetV2-SA, in which the activation function in top layers is replaced by ACON-C. Using information retrieval mechanism, M-Chatbot instructs patients to send breast pathological image to EfficientNetV2-SA network, and then the classifier trained by transfer learning will return the diagnosis results. We verify the performance of our chatbot and classification on the extrinsic metrics and BreaKHis dataset, respectively. The task completion rate of M-Chatbot reached 63.33\%. For the BreaKHis dataset, the highest accuracy of EfficientNetV2-SA network have achieved 84.71\%. All these experimental results illustrate that the proposed model can improve the accuracy performance of image recognition and our new intelligent medical diagnosis system is successful and efficient in providing automatic diagnosis of breast cancer.

cs.CV↗