arXiv subjects

Lu Yao

Publications and source records attributed to Lu Yao.

7 recordsLinked to original sources

arXiv2026

GI-Bench: A Panoramic Benchmark Revealing the Knowledge-Experience Dissociation of Multimodal Large Language Models in Gastrointestinal Endoscopy Against Clinical Standards

Multimodal Large Language Models (MLLMs) show promise in gastroenterology, yet their performance against comprehensive clinical workflows and human benchmarks remains unverified. To systematically evaluate state-of-the-art MLLMs across a panoramic gastrointestinal endoscopy workflow and determine their clinical utility compared with human endoscopists. We constructed GI-Bench, a benchmark encompassing 20 fine-grained lesion categories. Twelve MLLMs were evaluated across a five-stage clinical workflow: anatomical localization, lesion identification, diagnosis, findings description, and management. Model performance was benchmarked against three junior endoscopists and three residency trainees using Macro-F1, mean Intersection-over-Union (mIoU), and multi-dimensional Likert scale. Gemini-3-Pro achieved state-of-the-art performance. In diagnostic reasoning, top-tier models (Macro-F1 0.641) outperformed trainees (0.492) and rivaled junior endoscopists (0.727; p>0.05). However, a critical "spatial grounding bottleneck" persisted; human lesion localization (mIoU >0.506) significantly outperformed the best model (0.345; p<0.05). Furthermore, qualitative analysis revealed a "fluency-accuracy paradox": models generated reports with superior linguistic readability compared with humans (p<0.05) but exhibited significantly lower factual correctness (p<0.05) due to "over-interpretation" and hallucination of visual features. GI-Bench maintains a dynamic leaderboard that tracks the evolving performance of MLLMs in clinical endoscopy. The current rankings and benchmark results are available at https://roterdl.github.io/GIBench/.

cs.CV↗

arXiv2023

Single-pixel p-graded-n junction spectrometers

Ultra-compact spectrometers are becoming increasingly popular for their promising applications in biomedical analysis, environmental monitoring, and food safety. In this work, we report a novel single-pixel-photodetector spectrometer with a spectral range from 480 nm to 820 nm, based on the AlGaAs/GaAs p-graded-n junction with a voltage-tunable optical response. To reconstruct the optical spectrum, we propose a tailored method called Neural Spectral Fields (NSF) that leverages the unique wavelength and bias-dependent responsivity matrix. Our spectrometer achieves a high spectral wavelength accuracy of up to 0.30 nm and a spectral resolution of up to 10 nm. Additionally, we demonstrate the high spectral imaging performance of the device. The compatibility of our demonstration with the standard III-V process greatly accelerates the commercialization of miniaturized spectrometers.

physics.app-ph↗

arXiv2021

CrossFormer: A Versatile Vision Transformer Hinging on Cross-scale Attention

Transformers have made great progress in dealing with computer vision tasks. However, existing vision transformers do not yet possess the ability of building the interactions among features of different scales, which is perceptually important to visual inputs. The reasons are two-fold: (1) Input embeddings of each layer are equal-scale, so no cross-scale feature can be extracted; (2) to lower the computational cost, some vision transformers merge adjacent embeddings inside the self-attention module, thus sacrificing small-scale (fine-grained) features of the embeddings and also disabling the cross-scale interactions. To this end, we propose Cross-scale Embedding Layer (CEL) and Long Short Distance Attention (LSDA). On the one hand, CEL blends each embedding with multiple patches of different scales, providing the self-attention module itself with cross-scale features. On the other hand, LSDA splits the self-attention module into a short-distance one and a long-distance counterpart, which not only reduces the computational burden but also keeps both small-scale and large-scale features in the embeddings. Through the above two designs, we achieve cross-scale attention. Besides, we put forward a dynamic position bias for vision transformers to make the popular relative position bias apply to variable-sized images. Hinging on the cross-scale attention module, we construct a versatile vision architecture, dubbed CrossFormer, which accommodates variable-sized inputs. Extensive experiments show that CrossFormer outperforms the other vision transformers on image classification, object detection, instance segmentation, and semantic segmentation tasks. The code has been released: https://github.com/cheerss/CrossFormer.

cs.CV↗

arXiv2021

Context-aware Ensemble of Multifaceted Factorization Models for Recommendation Prediction in Social Networks

This paper describes the solution of Shanda Innovations team to Task 1 of KDD-Cup 2012. A novel approach called Multifaceted Factorization Models is proposed to incorporate a great variety of features in social networks. Social relationships and actions between users are integrated as implicit feedbacks to improve the recommendation accuracy. Keywords, tags, profiles, time and some other features are also utilized for modeling user interests. In addition, user behaviors are modeled from the durations of recommendation records. A context-aware ensemble framework is then applied to combine multiple predictors and produce final recommendation results. The proposed approach obtained 0.43959 (public score) / 0.41874 (private score) on the testing dataset, which achieved the 2nd place in the KDD-Cup competition.

cs.IR↗

arXiv2017

Curvature capillary repulsion

Directed assembly of colloids is an exciting field in materials science to form structures with new symmetries and responses. Fluid interfaces have been widely exploited to make densely packed ordered structures. We have been studying how interface curvature can be used in new ways to guide structure formation. On a fluid interface, the area of the deformation field around adsorbed microparticles depends on interface curvature; particles move to minimize the excess area of the distortions that they make in the interface. For particles that are sufficiently small, this area decreases as particles move along principle axes to sites of high deviatoric curvature. We have studied this migration for microparticles on a curved host interface with zero mean curvature created by pinning an oil-water interface around a micropost. Here, on a similar interface, we demonstrate capillary curvature repulsion, that is, we identify conditions in which microparticles migrate away from high curvature sites. Using theory and experiment, we discuss the origin of these interactions and their relationship to the particle's undulated contact line. We discuss the implications of this new type of interaction in various contexts from materials science to microrobotics.

cond-mat.soft↗

arXiv2014

Capillary migration of microdisks on curved interfaces

The capillary energy landscape for particles on curved fluid interfaces is strongly influenced by the particle wetting conditions. Contact line pinning has now been widely reported for colloidal particles, but its implications in capillary interactions have not been addressed. Here, we present experiment and analysis for disks with pinned contact lines on curved fluid interfaces. In experiment, we study microdisk migration on a host interface with zero mean curvature; the microdisks have contact lines pinned at their sharp edges and are sufficiently small that gravitational effects are negligible. The disks migrate away from planar regions toward regions of steep curvature with capillary energies inferred from the dissipation along particle trajectories which are linear in the deviatoric curvature. We derive the curvature capillary energy for an interface with arbitrary curvature, and discuss each contribution to the expression. By adsorbing to a curved interface, a particle eliminates a patch of fluid interface and perturbs the surrounding interface shape. Analysis predicts that perfectly smooth, circular disks do not migrate, and that nanometric deviations from a planar circular, contact line, like those around a weakly roughened planar disk, will drive migration with linear dependence on deviatoric curvature, in agreement with experiment.

cond-mat.soft↗

arXiv2013

Orbital stability of peakons for a generalized Camassa-Holm equation with both quadratic and cubic nonlinearity

In this paper, we investigate the orbital stability problem of peakons for a modified Camassa-Holm equation with both quadratic and cubic nonlinearity. This equation was derived from integrable theory and admits peaked soliton (peakon) and multipeakon solutions. By constructing two suitable piecewise functions, we establish the polynomial inequality relating to two conserved quantities and the maximum of the solution to this equation. The error estimate between the maximum of the solution and the peakon then follows from the structure of the polynomial inequality. Finally, we prove that a wave starting close to the peakon remains close to some translate of it at all later times, that is, the shapes of these peakons are stable under small perturbations.

math.AP↗