arXiv subjects

Yao Nie

Publications and source records attributed to Yao Nie.

16 recordsLinked to original sources

arXiv2026

When RAG Hurts: Diagnosing and Mitigating Attention Distraction in Retrieval-Augmented LVLMs

While Retrieval-Augmented Generation (RAG) is one of the dominant paradigms for enhancing Large Vision-Language Models (LVLMs) on knowledge-based VQA tasks, recent work attributes RAG failures to insufficient attention towards the retrieved context, proposing to reduce the attention allocated to image tokens. In this work, we identify a distinct failure mode that previous study overlooked: Attention Distraction (AD). When the retrieved context is sufficient (highly relevant or including the correct answer), the retrieved text suppresses the visual attention globally, and the attention on image tokens shifts away from question-relevant regions. This leads to failures on questions the model could originally answer correctly without the retrieved text. To mitigate this issue, we propose MAD-RAG, a training-free intervention that decouples visual grounding from context integration through a dual-question formulation, combined with attention mixing to preserve image-conditioned evidence. Extensive experiments on OK-VQA, E-VQA, and InfoSeek demonstrate that MAD-RAG consistently outperforms existing baselines across different model families, yielding absolute gains of up to 4.76%, 9.20%, and 6.18% over the vanilla RAG baseline. Notably, MAD-RAG rectifies up to 74.68% of failure cases with negligible computational overhead.

cs.CV↗

arXiv2026

Sharp non-uniqueness of weak solutions to 2D magnetohydrodynamic equations

In this paper, we prove that weak solutions to the 2D viscous and resistive magnetohydrodynamic (MHD) equations are non-unique in $L^2_t L^p(\mathbb{R}^2) \cap L^1_t W^{1,p}(\mathbb{R}^2)$ for given any $1\le p<\infty$, showing the sharpness of the Ladyzhenskaya--Prodi--Serrin condition at the endpoint $(2,\infty)$ and the solutions live on the borderline of the Beale--Kato--Majda criterion. To the best of our knowledge, this is the first non-uniqueness result for the 2D viscous and resistive MHD system. As byproducts, we also obtain non-uniqueness for the Navier--Stokes equations in $L^2_t L^p$ with $1\le p<\infty$, and for the MHD system with large $\mathrm{BMO}^{-1}$ initial data.

math.AP↗

arXiv2026

Visible Light Positioning With Lamé Curve LEDs: A Generic Approach for Camera Pose Estimation

Camera-based visible light positioning (VLP) is a promising technique for accurate and low-cost indoor camera pose estimation (CPE). To reduce the number of required light-emitting diodes (LEDs), advanced methods commonly exploit LED shape features for positioning. Although interesting, they are typically restricted to a single LED geometry, leading to failure in heterogeneous LED-shape scenarios. To address this challenge, this paper investigates Lamé curves as a unified representation of common LED shapes and proposes a generic VLP algorithm using Lamé curve-shaped LEDs, termed LC-VLP. In the considered system, multiple ceiling-mounted Lamé curve-shaped LEDs periodically broadcast their curve parameters via visible light communication, which are captured by a camera-equipped receiver. Based on the received LED images and curve parameters, the receiver can estimate the camera pose using LC-VLP. Specifically, an LED database is constructed offline to store the curve parameters, while online positioning is formulated as a nonlinear least-squares problem and solved iteratively. To provide a reliable initialization, a correspondence-free perspective-n-points (FreePnP) algorithm is further developed, enabling approximate CPE without any pre-calibrated reference points. The performance of LC-VLP is verified by both simulations and experiments. Simulations show that LC-VLP outperforms state-of-the-art methods in both circular- and rectangular-LED scenarios. Compared to a perspective arcs algorithm, LC-VLP can achieve reductions of both over 30% in average position and rotation errors. Experiments further show that LC-VLP can achieve an average position accuracy of less than 4 cm.

eess.SP↗

arXiv2026

Non-uniqueness of smooth solutions to the Navier-Stokes equations on torus $\TTT^2$

The local well-posedness theory for the incompressible Navier-Stokes equations in $\BMO^{-1}$ has attracted considerable attention over the past two decades. In a recent breakthrough, Coiculescu and Palasek (Invent. Math., 2025) settled the three-dimensional case by demonstrating the existence of two distinct global solutions, both smooth for $t>0$, evolving from a common initial datum in ${\rm BMO}^{-1}(\mathbb{T}^3)$. However, the two-dimensional case remains open. In this paper, we solve the two-dimensional problem. Unlike its three-dimensional counterpart, the two-dimensional setting presents additional difficulties stemming from the geometric intersections of two-dimensional Mikado flows. To overcome these difficulties, we develop a heat-dominated Fourier mode flow built upon steady two-dimensional Euler flows, and present the proof using a new iterative scheme.

math.AP↗

arXiv2025

SlideMamba: Entropy-Based Adaptive Fusion of GNN and Mamba for Enhanced Representation Learning in Digital Pathology

Advances in computational pathology increasingly rely on extracting meaningful representations from Whole Slide Images (WSIs) to support various clinical and biological tasks. In this study, we propose a generalizable deep learning framework that integrates the Mamba architecture with Graph Neural Networks (GNNs) for enhanced WSI analysis. Our method is designed to capture both local spatial relationships and long-range contextual dependencies, offering a flexible architecture for digital pathology analysis. Mamba modules excels in capturing long-range global dependencies, while GNNs emphasize fine-grained short-range spatial interactions. To effectively combine these complementary signals, we introduce an adaptive fusion strategy that uses an entropy-based confidence weighting mechanism. This approach dynamically balances contributions from both branches by assigning higher weight to the branch with more confident (lower-entropy) predictions, depending on the contextual importance of local versus global information for different downstream tasks. We demonstrate the utility of our approach on a representative task: predicting gene fusion and mutation status from WSIs. Our framework, SlideMamba, achieves an area under the precision recall curve (PRAUC) of 0.751 \pm 0.05, outperforming MIL (0.491 \pm 0.042), Trans-MIL (0.39 \pm 0.017), Mamba-only (0.664 \pm 0.063), GNN-only (0.748 \pm 0.091), and a prior similar work GAT-Mamba (0.703 \pm 0.075). SlideMamba also achieves competitive results across ROC AUC (0.738 \pm 0.055), sensitivity (0.662 \pm 0.083), and specificity (0.725 \pm 0.094). These results highlight the strength of the integrated architecture, enhanced by the proposed entropy-based adaptive fusion strategy, and suggest promising potential for application of spatially-resolved predictive modeling tasks in computational pathology.

cs.CV↗

arXiv2025

On Onsager-type conjecture for the Elsässer energies of the ideal MHD equations

In this paper, we investigate the ideal magnetohydrodynamics (MHD) equations on tours $\TTT^d$. For $d=3$, we resolve the flexible part of Onsager-type conjecture for Elsässer energies of the ideal MHD equations. More precisely, for $β< 1/3$, we construct weak solutions $(u, b) \in C^β([0,T] \times \mathbb{T}^3)$ with both the total energy dissipation and failure of cross helicity conservation. The key idea of the proof relies on a symmetry reduction that embeds the ideal MHD system into a 2$\frac{1}{2}$D Euler flow and the Newton-Nash iteration technique recently developed in \cite{GR}. For $d=2$, we show the non-uniqueness of Hölder-continuous weak solutions with non-trivial magnetic fields. Specifically, for $β< 1/5$, there exist infinitely many solutions $(u, b) \in C^β([0,T] \times \mathbb{T}^2)$ with the same initial data while satisfying the total energy dissipation with non-vanishing velocity and magnetic fields. The new ingredient is developing a spatial-separation-driven iterative scheme that incorporates the magnetic field as a controlled perturbation within the convex integration framework for the velocity field, thereby providing sufficient oscillatory freedom for Nash-type perturbations in the 2D setting. As a byproduct, we prove that any Hölder-continuous Euler solution can be approximated by a sequence of $C^β$-weak solutions for the ideal MHD equations in the $L^p$-topology for $1\le p<\infty$.

math.AP↗

arXiv2024

Sharp non-uniqueness for the Navier-Stokes equations in R^3

In this paper, we prove a sharp and strong non-uniqueness for a class of weak solutions to the incompressible Navier-Stokes equations in $\R^3$. To be more precise, we exhibit the non-uniqueness result in a strong sense, that is, any weak solution is non-unique in L^p([0,T];L^\infty(\R^3)) with 1\le p<2. Moreover, this non-uniqueness result is sharp with regard to the classical Ladyzhenskaya-Prodi-Serrin criteria at endpoint (2, \infty), which extends the sharp nonuniqueness for the Navier-Stokes equations on torus $\TTT^3$ in the recent groundbreaking work (Cheskidov and Luo, Invent. Math., 229 (2022), pp. 987-1054) to the setting of the whole space. The key ingredient is developing a new iterative scheme that balances the compact support of the Reynolds stress error with the non-compact support of the solution via introducing incompressible perturbation fluid.

math.AP↗

arXiv2024

Non-uniqueness of weak solutions to the Navier-Stokes equations in R^3

To our knowledge, the convex integration method has been widely applied to the study of non-uniqueness of solutions to the Naiver-Stokes equations in the periodic region, but there are few works on applying this method to the corresponding problems in the whole space or other regions. In this paper, we prove that weak solutions of the Navier-Stokes equations are not unique in the class of weak solutions with finite kinetic energy in the whole space, which extends the non uniqueness result for the Navier-Stokes equations on torus T3in the groundbreaking work (Buckmaster and Vicol, Ann. of Math., 189 (2019), pp.101-144) to R3. The critical ingredients of the proof include developing an iterative scheme in which the approximation solution is refined by decomposing it into local and non-local parts. For the non-local part, we introduce the localized corrector which plays a crucial role in balancing the compact support of the Reynolds stress error with the non-compact support of the solution. As applications of this argument, we first prove that there exist infinitely many weak solutions that dissipate the kinetic energy in smooth bounded domain. Moreover, we show the instability of the Navier-Stokes equations near Couette flow in L2(R3).

math.AP↗

arXiv2024

A Novel Generative Artificial Intelligence Method for Interference Study on Multiplex Brightfield Immunohistochemistry Images

Multiplex brightfield imaging offers the advantage of simultaneously analyzing multiple biomarkers on a single slide, as opposed to single biomarker labeling on multiple consecutive slides. To accurately analyze multiple biomarkers localized at the same cellular compartment, two representative biomarker sets were selected as assay models - cMET-PDL1-EGFR and CD8-LAG3-PDL1, where all three biomarkers can co-localize on the cell membrane. One of the most crucial preliminary stages for analyzing such assay is identifying each unique chromogen on individual cells. This is a challenging problem due to the co-localization of membrane stains from all the three biomarkers. It requires advanced color unmixing for creating the equivalent singleplex images from each triplex image for each biomarker. In this project, we developed a cycle-Generative Adversarial Network (cycle-GAN) method for unmixing the triplex images generated from the above-mentioned assays. Three different models were designed to generate the singleplex image for each of the three stains Tamra (purple), QM-Dabsyl (yellow) and Green. A notable novelty of our approach was that the input to the network were images in the optical density domain instead of conventionally used RGB images. The use of the optical density domain helped in reducing the blurriness of the synthetic singleplex images, which was often observed when the network was trained on RGB images. The cycle-GAN models were validated on 10,800 lung, gastric and colon images for the cMET-PDL1-EGFR assay and 3600 colon images for the CD8-LAG3-PDL1 assay. Visual as well as quantified assessments demonstrated that the proposed method is effective and efficient when compared with the manual reviewing results and is readily applicable to various multiplex assays.

eess.IV↗

arXiv2024

On Onsager's type conjecture for the inviscid Boussinesq equations

In this paper, we investigate the Cauchy problem for the three dimensional inviscid Boussinesq system in the periodic setting. For $1\le p\le \infty$, we show that the threshold regularity exponent for $L^p$-norm conservation of temperature of this system is $1/3$, consistent with Onsager exponent. More precisely, for $1\le p\le\infty$, every weak solution $(v,θ)\in C_tC^β_x$ to the inviscid Boussinesq equations satisfies that $\|θ(t)\|_{L^p(\mathbb{T}^3)}=\|θ_0\|_{L^p(\mathbb{T}^3)}$ if $β>\frac{1}{3}$, while if $β<\frac{1}{3}$, there exist infinitely many weak solutions $(v,θ)\in C_tC^β_x$ such that the $L^p$-norm of temperature is not conserved. As a byproduct, we are able to construct many weak solutions in $C_tC^β_x$ for $β<\frac{1}{3}$ displaying wild behavior, such as fast kinetic energy dissipation and high oscillation of velocity. Moreover, we also show that if a weak solution $(v, θ)$ of this system has at least one interval of regularity, then this weak solution $(v,θ)$ is not unique in $C_tC^β_x$ for $β<\frac{1}{3}$.

math.AP↗

arXiv2024

Sharp ill-posedness for the non-resistive MHD equations in Sobolev spaces

In this paper, we prove a sharp ill-posedness result for the incompressible non-resistive MHD equations. In any dimension $d\ge 2$, we show the ill-posedness of the non-resistive MHD equations in $H^{\frac{d}{2}-1}(\mathbb{R}^d)\times H^{\frac{d}{2}}(\mathbb{R}^d)$, which is sharp in view of the results of the local well-posedness in $H^{s-1}(\mathbb{R}^d)\times H^{s}(\mathbb{R}^d)(s>\frac{d}{2})$ established by Fefferman et al.(Arch. Ration. Mech. Anal., \textbf{223} (2), 677-691, 2017). Furthermore, we generalize the ill-posedness results from $H^{\frac{d}{2}-1}(\mathbb{R}^d)\times H^{\frac{d}{2}}(\mathbb{R}^d)$ to Besov spaces $B^{\frac{d}{p}-1}_{p, q}(\mathbb{R}^d)\times B^{\frac{d}{p}}_{p, q}(\mathbb{R}^d)$ and $\dot B^{\frac{d}{p}-1}_{p, q}(\mathbb{R}^d)\times \dot B^{\frac{d}{p}}_{p, q}(\mathbb{R}^d)$ for $1\le p\le\infty, q>1$. Different from the ill-posedness mechanism of the incompressible Navier-Stokes equations in $\dot B^{-1}_{\infty, q}$ \cite{B,W}, we construct an initial data such that the paraproduct terms (low-high frequency interaction) of the nonlinear term make the main contribution to the norm inflation of the magnetic field.

math.AP↗

arXiv2024

Cross-modality Attention-based Multimodal Fusion for Non-small Cell Lung Cancer (NSCLC) Patient Survival Prediction

Cancer prognosis and survival outcome predictions are crucial for therapeutic response estimation and for stratifying patients into various treatment groups. Medical domains concerned with cancer prognosis are abundant with multiple modalities, including pathological image data and non-image data such as genomic information. To date, multimodal learning has shown potential to enhance clinical prediction model performance by extracting and aggregating information from different modalities of the same subject. This approach could outperform single modality learning, thus improving computer-aided diagnosis and prognosis in numerous medical applications. In this work, we propose a cross-modality attention-based multimodal fusion pipeline designed to integrate modality-specific knowledge for patient survival prediction in non-small cell lung cancer (NSCLC). Instead of merely concatenating or summing up the features from different modalities, our method gauges the importance of each modality for feature fusion with cross-modality relationship when infusing the multimodal features. Compared with single modality, which achieved c-index of 0.5772 and 0.5885 using solely tissue image data or RNA-seq data, respectively, the proposed fusion approach achieved c-index 0.6587 in our experiment, showcasing the capability of assimilating modality-specific knowledge from varied modalities.

eess.IV↗

arXiv2022

Sharp and strong non-uniqueness for the magneto-hydrodynamic equations

In this paper, we prove a sharp and strong non-uniqueness for a class of weak solutions to the three-dimensional magneto-hydrodynamic (MHD) system. More precisely, we show that any weak solution $(v,b)\in L^p_tL^{\infty}_x$ is non-unique in $L^p_tL^{\infty}_x$ with $1\le p<2$, which reveals the strong non-uniqueness, and the sharpness in terms of the classical Ladyzhenskaya-Prodi-Serrin criteria at endpoint $(2, \infty)$. Moreover, for any $1\le p<2$ and $ε>0$, we construct non-Leray-Hopf weak solutions in $L^p_tL^{\infty}_x\cap L^1_tC^{1-ε}$. The results of Navier-Stokes equations in \cite{1Cheskidov} imply the sharp non-uniqueness of MHD system with trivial magnetic field $b$. Our result shows the non-uniqueness for any weak solution $(v,b)$ including non-trivial magnetic field $b$.

math.AP↗

arXiv2022

Ill-posedness issue for the 2D viscous shallow water equations in some critical Besov spaces

We study the Cauchy problem of the 2D viscous shallow water equations in some critical Besov spaces $\dot B^{\frac{2}{p}}_{p,1}(\mathbb{R}^2)\times \dot B^{\frac{2}{p}-1}_{p,q}(\mathbb{R}^2)$. As is known, this system is locally well-posed for large initial data as well as globally well-posed for small initial data in $\dot B^{\frac{2}{p}}_{p,1}(\mathbb{R}^2)\times \dot B^{\frac{2}{p}-1}_{p,1}(\mathbb{R}^2)$ for $p<4$ and ill-posed in $\dot B^{\frac{2}{p}}_{p,1}(\mathbb{R}^2)\times \dot B^{\frac{2}{p}-1}_{p,1}(\mathbb{R}^2)$ for $p>4$. In this paper, we prove that this system is ill-posed for the critical case $p=4$ in the sense of "norm inflation". Furthermore, we also show that the system is ill-posed in $\dot B^{\frac{1}{2}}_{4,1}(\mathbb{R}^2)\times \dot B^{-\frac{1}{2}}_{4,q}(\mathbb{R}^2)$ for any $q\neq 2$

math.AP↗

arXiv2022

Global-in-time Boundedness of solution for Cauchy problem to the Parabolic-Parabolic Keller-Segel system with logistic growth

We study global-in-time well-posedness and the behaviour and of the solution to Cauchy problem in the classical Keller-Segel system with logistic term \begin{equation*} \left. \aligned \partial_tn-Δn=&-χ\nabla\cdot(n\nabla c)+\la n-μn^2 τ\partial_tc-Δc=&-c+n \endaligned \right\}\quad\text{in}\,\,\,\RR^d\times\RR^+, \end{equation*} where $d\ge 1$, $τ,\, χ,\, μ>0$ and $λ\ge 0$. It's inspired by a previous result \cite[M. Winkler, Commun. Part. Diff. Eq., 35 (2010), 1516-1537]{Win10}, where the global-in-time boundedness of the above Keller-Segel system in smooth \emph{bounded }convex domains is established for large $μ$. However, his approach in bounded domain ceases to directly apply in the entire space $\RR^d$, and then they raised an interesting question whether a similar global-in-time boundedness statement remains true of Cauchy problem. In this paper, we answer this open problem by developing local-in-space estimates. More precisely, we prove that the above Keller-Segel system possesses a uniquely global-in-time bounded solution for any $τ>0$ under the assumption that $μ$ is large. The key point of our proof heavily relies on localization in space of solution caused by "local effect" of $L^\infty(\RR^d)$-norm.

math.AP↗

arXiv2019

Weakly Supervised Multi-Task Learning for Cell Detection and Segmentation

Cell detection and segmentation is fundamental for all downstream analysis of digital pathology images. However, obtaining the pixel-level ground truth for single cell segmentation is extremely labor intensive. To overcome this challenge, we developed an end-to-end deep learning algorithm to perform both single cell detection and segmentation using only point labels. This is achieved through the combination of different task orientated point label encoding methods and a multi-task scheduler for training. We apply and validate our algorithm on PMS2 stained colon rectal cancer and tonsil tissue images. Compared to the state-of-the-art, our algorithm shows significant improvement in cell detection and segmentation without increasing the annotation efforts.

cs.CV↗