arXiv subjects

Yulong Zheng

Publications and source records attributed to Yulong Zheng.

10 recordsLinked to original sources

arXiv2026

BehaviorWorldGen: Closing the Loop between Action Models and World Simulators via Controllable Behavior-Aware Structured World Generation

Modern driving action models are increasingly improved in a self-improvement loop, where a learned world simulator imagines future observations and the resulting data is fed back to refine the action model. However, the bottleneck of this loop lies in the simulators' inability to generate behaviorally plausible responses by surrounding agents, making generated data both unrealistic in interaction and imbalanced in distribution. We introduce BehaviorWorldGen, a framework that closes the loop between action models and world simulators through controllable behavior-aware structured world generation. Its core component is BehaviorFlow, a meta-action-conditioned traffic-flow model that injects interpretable behavior controls and jointly generates multi-agent rollouts. BehaviorFlow realizes the specified agent behaviors while allowing surrounding vehicles to respond to the ego and to one another. The resulting rollouts are rendered by a world simulator into realistic multi-view observations, which are paired with corrected interaction-aware trajectories for action-model refinement. Since BehaviorWorldGen uses structured trajectories as the interface between its modules, it is compatible with diverse action models and world simulators. Experiments on world generation, scene extrapolation, and policy refinement demonstrate consistent improvements, with the largest benefits concentrated on difficult interactive scenarios.

cs.RO↗

arXiv2026

Adversarial LassoNet: Robust Feature Selection via Stability-Driven Sparse Learning

Sparse feature selection is critical for high-dimensional machine learning, yet traditional $\ell_1$-regularized methods are often brittle under observational noise and spurious correlations, leading to unstable feature supports and degraded generalization. Although adversarial training has been widely used to improve model robustness, its interaction with hierarchical sparse feature selection remains underexplored. In this work, we propose Adversarial LassoNet (AdLNet), a stability-driven sparse feature selection framework that integrates input-space adversarial perturbations with the hierarchical sparsity mechanism of LassoNet. We derive a tractable first-order adversarial approximation under local smoothness assumptions and provide an NTK-inspired spectral analysis to characterize how perturbation-driven training can reduce gradient concentration. Experiments on high-dimensional SERS data, six public benchmark datasets, and ColoredMNIST show that AdLNet maintains competitive sparse-selection performance while improving out-of-distribution robustness by 4.4\% and feature support reproducibility by 6.3\% under nearly matched support sparsity on ColoredMNIST. On the high-dimensional lung cancer screening dataset, AdLNet achieves a 5.3\% test accuracy gain and a 6.0\% AUC improvement over vanilla LassoNet. Code and dataset are available at https://github.com/719573/Adversarial-LassoNet.

cs.LG↗

arXiv2025

pH Regulates Ion Dynamics in Carboxylated Mixed Conductors

Coupled ionic and electronic transport underpins processes as diverse as electrochemical energy conversion, biological signaling, and soft adaptive electronics. Yet, how chemical environments such as pH modulate this coupling at the molecular scale remains poorly understood. Here, we show that the protonation state of carboxylated polythiophenes provides precise chemical control over ion dynamics, doping efficiency, solvent uptake and mechanical response. The findings establish molecular acidity as a general strategy to program ionic preference and mechanical stability, offering design principles for pH-responsive mixed conductors and soft electronic materials that couple ionic, electronic, and mechanical functionality.

cond-mat.mtrl-sci↗

arXiv2025

AvatarVTON: 4D Virtual Try-On for Animatable Avatars

We propose AvatarVTON, the first 4D virtual try-on framework that generates realistic try-on results from a single in-shop garment image, enabling free pose control, novel-view rendering, and diverse garment choices. Unlike existing methods, AvatarVTON supports dynamic garment interactions under single-view supervision, without relying on multi-view garment captures or physics priors. The framework consists of two key modules: (1) a Reciprocal Flow Rectifier, a prior-free optical-flow correction strategy that stabilizes avatar fitting and ensures temporal coherence; and (2) a Non-Linear Deformer, which decomposes Gaussian maps into view-pose-invariant and view-pose-specific components, enabling adaptive, non-linear garment deformations. To establish a benchmark for 4D virtual try-on, we extend existing baselines with unified modules for fair qualitative and quantitative comparisons. Extensive experiments show that AvatarVTON achieves high fidelity, diversity, and dynamic garment realism, making it well-suited for AR/VR, gaming, and digital-human applications.

cs.CV↗

arXiv2025

NexusGS: Sparse View Synthesis with Epipolar Depth Priors in 3D Gaussian Splatting

Neural Radiance Field (NeRF) and 3D Gaussian Splatting (3DGS) have noticeably advanced photo-realistic novel view synthesis using images from densely spaced camera viewpoints. However, these methods struggle in few-shot scenarios due to limited supervision. In this paper, we present NexusGS, a 3DGS-based approach that enhances novel view synthesis from sparse-view images by directly embedding depth information into point clouds, without relying on complex manual regularizations. Exploiting the inherent epipolar geometry of 3DGS, our method introduces a novel point cloud densification strategy that initializes 3DGS with a dense point cloud, reducing randomness in point placement while preventing over-smoothing and overfitting. Specifically, NexusGS comprises three key steps: Epipolar Depth Nexus, Flow-Resilient Depth Blending, and Flow-Filtered Depth Pruning. These steps leverage optical flow and camera poses to compute accurate depth maps, while mitigating the inaccuracies often associated with optical flow. By incorporating epipolar depth priors, NexusGS ensures reliable dense point cloud coverage and supports stable 3DGS training under sparse-view conditions. Experiments demonstrate that NexusGS significantly enhances depth accuracy and rendering quality, surpassing state-of-the-art methods by a considerable margin. Furthermore, we validate the superiority of our generated point clouds by substantially boosting the performance of competing methods. Project page: https://usmizuki.github.io/NexusGS/.

cs.CV↗

arXiv2024

Unveiling multi-quantum excitonic correlations in push-pull polymer semiconductors

Bound and unbound Frenkel-exciton pairs are essential transient precursors for a variety of photophysical and biochemical processes. In this work, we identify bound and unbound {Frenkel}-exciton complexes in an electron push-pull polymer semiconductor using coherent two-dimensional spectroscopy. We find that the dominant $A_{0-1}$ peak of the absorption vibronic progression is accompanied by a sub-peak, each dressed by distinct vibrational modes. By considering the Liouville pathways within a two-exciton model, the imbalanced cross peaks in one-quantum rephasing and non-rephasing spectra can be accounted for by the presence of pure biexcitons. The two-quantum non-rephasing spectra, on the other hand, provide direct evidence for unbound exciton pairs and biexcitons with dominantly attractive force. In addition, the spectral features of unbound exciton pairs show mixed absorptive and dispersive character, implying many-body interactions within the correlated {Frenkel}-exciton pairs. Our work offers novel perspectives on the rich photophysical processes in semiconductor polymers with the presence of Frenkel exciton complexes.

cond-mat.mtrl-sci↗

arXiv2023

Exciton Bimolecular Annihilation Dynamics in Push-Pull Semiconductor Polymers

Exciton-exciton annihilation is a ubiquitous nonlinear dynamical phenomenon in materials hosting Frenkel excitons, that has been employed to probe exciton diffusion processes in conjugated polymeric materials. In this work, we investigate the nonlinear exciton dynamics of an electron push-pull conjugated polymer by fluence-dependent transient absorption and excitation-correlation photoluminescence spectroscopy, where we can quantitatively show the latter technique to be a more selective probe of the nonlinear dynamics. Simulations based on an exciton annihilation model that implements a simple (\textit{i.e.}\ time-independent) bimolecular rate constant decreasing trend for the extracted annihilation rates with excitation fluence. However, further investigation of the fluence-dependent transients suggests that the exciton-exciton annihilation bimolecular rate parameters are not constant in time, displaying a $t^{-1/2}$ time dependence, which we rationalize as reflective of anisotropic exciton diffusion. At ambient temperature, we estimate the exciton diffusion length to be $9 \pm 2$\,nm. In addition, exciton annihilation gives rise to a long-lived species that recombines on a nanosecond timescale. Our conclusions shed broad light onto nonlinear exciton dynamics in push-pull conjugated polymers.

cond-mat.mtrl-sci↗

arXiv2023

Parameter Estimation of LAMOST Medium-resolution Stellar Spectra

This paper investigates the problem of estimating three stellar atmospheric physical parameters and thirteen elemental abundances for medium-resolution spectra from Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST). Typical characteristics of these spectra are their huge scale, wide range of spectral signal-to-noise ratios, and uneven distribution in parameter space.These characteristics lead to unsatisfactory results on the spectra with low temperature, high temperature or low metallicity.To this end, this paper proposes a Stellar Parameter Estimation method based on Multiple Regions (SPEMR) that effectively improves parameter estimation accuracy. On the spectra with {S/N $\geq 10$}, the precisions are 47 K, 0.08 dex, 0.03 dex respectively for the estimations of ($T_{\rm eff}$, $\log \,g$ and $\rm [Fe/H]$), 0.03 dex to 0.06 dex for elements C, Mg, Al, Si, Ca, Mn and Ni, 0.07 dex to 0.13 dex for N, O, S, K and Ti, while that of Cr is 0.16 dex.For the reference of astronomical science researchers and algorithm researchers, we released a catalog for 4.19 million medium-resolution spectra from the LAMOST DR8, experimental code, trained model, training data, and test data.

astro-ph.SR↗

arXiv2023

Chain Conformation and Exciton Delocalization in a Push-Pull Conjugated Polymer

Linear and nonlinear optical lineshapes reveal details of excitonic structure in semiconductor polymers. We implement absorption, photoluminescence, and transient absorption spectroscopies in DPP-DTT, an electron push-pull copolymer, to explore the relationship between their spectral lineshapes and chain conformation, deduced from resonance Raman spectroscopy and from \textit{ab initio} calculations. The viscosity of precursor polymer solutions before film casting displays a transition that suggests gel formation above a critical concentration. Upon crossing this viscosity deflection concentration, the lineshape analysis of the absorption spectra within a photophysical aggregate model reveals a gradual increase in interchain excitonic coupling. We also observe a red-shifted and line-narrowed steady-state photoluminescence spectrum, along with increasing resonance Raman intensity in the stretching and torsional modes of the dithienothiphene unit, which suggests a longer exciton coherence length along the polymer-chain backbone. Furthermore, we observe a change of lineshape in the photoinduced absorption component of the transient absorption spectrum. The derivative-like lineshape may originate from two possibilities: a new excited-state absorption, or from optical Stark effect, both of which are consistent with the emergence of high-energy shoulder as seen in both photoluminescence and absorption spectra. Therefore, we conclude that the exciton is more dispersed along the polymer chain backbone with increasing concentrations, leading to the hypothesis that the polymer chain order is enhanced when the push-pull polymers are processed at higher concentrations. Thus, tuning the microscopic chain conformation by concentration would be another factor of interest when considering the polymer assembly pathways for pursuing large-area and high-performance organic optoelectronic devices.

cond-mat.mtrl-sci↗

arXiv2023

Resolving nonlinear recombination dynamics in semiconductors via ultrafast excitation correlation spectroscopy: Photoluminescence versus photocurrent detection

We explore the application of excitation correlation spectroscopy to detect nonlinear photophysical dynamics in two distinct semiconductor classes through time-integrated photoluminescence and photocurrent measurements. In this experiment, two variably delayed femtosecond pulses excite the semiconductor, and the time-integrated photoluminescence or photocurrent component arising from the nonlinear dynamics of the populations induced by each pulse is measured as a function of inter-pulse delay by phase-sensitive detection with a lock-in amplifier. We focus on two limiting materials systems with contrasting optical properties: a prototypical lead-halide perovskite (LHP) solar cell, in which primary photoexcitations are charge photocarriers, and a single-component organic-semiconductor diode, which features Frenkel excitons as primary photoexcitations. The photoexcitation dynamics perceived by the two detection schemes in these contrasting systems are distinct. Nonlinear-dynamic contributions in the photoluminescence detection scheme arise from contributions to radiative recombination in both materials systems, while photocurrent arises directly in the LHP but indirectly following exciton dissociation in the organic system. Consequently, the basic photophysics of the two systems are reflected differently when comparing measurements with the two detection schemes.

cond-mat.mtrl-sci↗