arXiv subjects

Xiaodong Yuan

Publications and source records attributed to Xiaodong Yuan.

6 recordsLinked to original sources

arXiv2026

Metadata-Free Meta-Reweighted Direct Preference Optimization under Noisy Preference Labels

Direct Preference Optimization (DPO) has become an important method for aligning large language models (LLMs) with human preferences because it removes the need for explicit reward modeling and reinforcement learning. However, its performance depends heavily on the quality of preference data, and noisy preference data in real-world settings can weaken alignment performance. To address this issue, we propose a bilevel optimization framework and prove, under some idealized conditions, that this framework can recover the DPO optimum under clean data. We further derive a prior form for the learnable weighting function under label-flipping noise. Considering that high-quality metadata may be difficult to obtain, we propose a prompt augmentation consistency method that enables meta-learning even when metadata is completely unavailable. To reduce the high cost of higher-order gradients in LLM meta-learning, we combine central-difference approximation with LoRA fine-tuning and develop a scalable training scheme. Experiments on TL;DR summarization and Anthropic Helpful and Harmless dialogue show that the proposed method improves alignment performance over multiple DPO baselines under different noise rates.

cs.LG↗

arXiv2021

Sleep Staging Based on Multi Scale Dual Attention Network

Sleep staging plays an important role on the diagnosis of sleep disorders. In general, experts classify sleep stages manually based on polysomnography (PSG), which is quite time-consuming. Meanwhile, the acquisition process of multiple signals is much complex, which can affect the subject's sleep. Therefore, the use of single-channel electroencephalogram (EEG) for automatic sleep staging has become a popular research topic. In the literature, a large number of sleep staging methods based on single-channel EEG have been proposed with promising results and achieve the preliminary automation of sleep staging. However, the performance for most of these methods in the N1 stage do not satisfy the needs of the diagnosis. In this paper, we propose a deep learning model multi scale dual attention network(MSDAN) based on raw EEG, which utilizes multi-scale convolution to extract features in different waveforms contained in the EEG signal, connects channel attention and spatial attention mechanisms in series to filter and highlight key information, and uses soft thresholding to remove redundant information. Experiments were conducted using two datasets with 5-fold cross-validation and hold-out validation method. The final average accuracy, overall accuracy, macro F1 score and Cohen's Kappa coefficient of the model reach 96.70%, 91.74%, 0.8231 and 0.8723 on the Sleep-EDF dataset, 96.14%, 90.35%, 0.7945 and 0.8284 on the Sleep-EDFx dataset. Significantly, our model performed superiorly in the N1 stage, with F1 scores of 54.41% and 52.79% on the two datasets respectively. The results show the superiority of our network over the existing methods, reaching a new state-of-the-art. In particular, the proposed method achieves excellent results in the N1 sleep stage compared to other methods.

cs.LG↗

arXiv2018

Large Chiroptical Effects in Planar Chiral Metamaterials

Chiroptical effects, characterized by different optical responses for left- (LCP) and right- handed circularly polarized light (RCP), are powerful and valuable tools in optics with wide applications in polarization resolved imaging and sensing. Previously observed strong chiroptical effects are limited to metamaterials with complex three-dimensional chiral structures at the sub-wavelength scale. Although asymmetrical transmission of LCP and RCP have been investigated in planar chiral metasurfaces, the observed weak chiroptical effects result from anisotropic Ohmic dissipation of the metal constituents. Here, we demonstrate by theory and proof-of-concept experiments that large difference in transmittances of LCP and RCP can be attained in a single-layer planar chiral metamaterial with a sub-wavelength thickness. Without violating the reciprocity and mirror symmetry, the strong chiroptical effect, independent of dielectric loss, arise from a novel mechanism of multimode interference. The described effect may lead to new gateway towards chiral manipulations of light and chiral optical devices.

physics.optics↗

arXiv2015

Graphene plasmonics for light trapping and absorption engineering

Plasmonics can be used to improve absorption in optoelectronic devices and has been intensively studied for solar cells and photodetectors. Graphene has recently emerged as a powerful plasmonic material. It shows significantly less losses compared to traditional plasmonic materials such as gold and silver and its plasmons can be tuned by changing the Fermi energy with chemical or electrical doping. Here we propose the usage of graphene plasmonics for light trapping in optoelectronic devices and show that the excitation of localized plasmons in doped, nanostructured graphene can enhance optical absorption in its surrounding media including both bulky and two-dimensional materials by tens of times, which may lead to a new generation of highly efficient, spectrally selective photodetectors in mid-infrared and THz ranges. The proposed concept could even revolutionize the field of plasmonic solar cells if graphene plasmons in the visible and near-infrared are realized.

physics.optics↗

arXiv2015

Tunable asymmetric transmission of THz wave through a graphene planar chiral structure

In this letter, we show that asymmetric transmission of circularly polarized waves through a nanostructured planar chiral graphene film can be observed in terahertz range. The asymmetric transmission effect of monatomic layer graphene closely resembles that of metallic planar chiral nanostructures which has previously been demonstrated. And the relative enantiomeric difference in the total transmission varies with the change of graphene's Fermi level. The plasmonic excitation in the graphene nanostructure is the enantiometically sensitive which is asymmetric for opposite propagating directions. This phenomenon will deepen our understanding of light-matter interactions in planar chiral structures and may find applications in polarization-sensitive devices, sensors, detectors and other areas.

physics.optics↗

arXiv2014

Circular Polarizer Realized by a Single Layer of Planar Chiral Metallic Nanostructure

As a basic optical element, circular polarizer plays significant roles in signal transmission, measurements and life science microscopy. Three-dimensional (3D) chiral structures have been thought to be necessary to realize circular polarizers. Here we demonstrate theoretically and experimentally for the first time that a high-efficiency circular polarizer could be realized by a single layer of planar 2D chiral structure. Our proposed circular polarizer is based on unidirectional polarization conversion instead of circular polarization stop bands. Since two-dimensional planar structures present obvious advantage for fabrication and integration on chip, the proposed circular polarizer is of great interest in integrated optics and microscopy. It provides a novel scheme to manipulate polarizations of light wave, as well as Terahertz wave and microwave.

physics.optics↗