arXiv subjects

Thinh Tien Nguyen

Publications and source records attributed to Thinh Tien Nguyen.

3 recordsLinked to original sources

arXiv2024

Semi-Supervised Semantic Segmentation using Redesigned Self-Training for White Blood Cells

Artificial Intelligence (AI) in healthcare, especially in white blood cell cancer diagnosis, is hindered by two primary challenges: the lack of large-scale labeled datasets for white blood cell (WBC) segmentation and outdated segmentation methods. These challenges inhibit the development of more accurate and modern techniques to diagnose cancer relating to white blood cells. To address the first challenge, a semi-supervised learning framework should be devised to efficiently capitalize on the scarcity of the dataset available. In this work, we address this issue by proposing a novel self-training pipeline with the incorporation of FixMatch. Self-training is a technique that utilizes the model trained on labeled data to generate pseudo-labels for the unlabeled data and then re-train on both of them. FixMatch is a consistency-regularization algorithm to enforce the model's robustness against variations in the input image. We discover that by incorporating FixMatch in the self-training pipeline, the performance improves in the majority of cases. Our performance achieved the best performance with the self-training scheme with consistency on DeepLab-V3 architecture and ResNet-50, reaching 90.69%, 87.37%, and 76.49% on Zheng 1, Zheng 2, and LISC datasets, respectively.

cs.CV↗

arXiv2017

Asymptotic limit and decay estimates for a class of dissipative linear hyperbolic systems in several dimensions

In this paper, we study the large-time behavior of solutions to a class of partially dissipative linear hyperbolic systems with applications in velocity-jump processes in several dimensions. Given integers $n,d\ge 1$, let $\mathbf A:=(A^1,\dots,A^d)\in (\mathbb R^{n\times n})^d$ be a matrix-vector, where $A^j\in\mathbb R^{n\times n}$, and let $B\in \mathbb R^{n\times n}$ be not required to be symmetric but have one single eigenvalue zero, we consider the Cauchy problem for linear $n\times n$ systems having the form \begin{equation*} \partial_{t}u+\mathbf A\cdot \nabla_{\mathbf x} u+Bu=0,\qquad (\mathbf x,t)\in \mathbb R^d\times \mathbb R_+. \end{equation*} Under appropriate assumptions, we show that the solution $u$ is decomposed into $u=u^{(1)}+u^{(2)}$, where $u^{(1)}$ has the asymptotic profile which is the solution, denoted by $U$, of a parabolic equation and $u^{(1)}-U$ decays at the rate $t^{-\frac d2(\frac 1q-\frac 1p)-\frac 12}$ as $t\to +\infty$ in any $L^p$-norm, and $u^{(2)}$ decays exponentially in $L^2$-norm, provided $u(\cdot,0)\in L^q(\mathbb R^d)\cap L^2(\mathbb R^d)$ for $1\le q\le p\le \infty$. Moreover, $u^{(1)}-U$ decays at the optimal rate $t^{-\frac d2(\frac 1q-\frac 1p)-1}$ as $t\to +\infty$ if the system satisfies a symmetry property. The main proofs are based on asymptotic expansions of the solution $u$ in the frequency space and the Fourier analysis.

math.AP↗

arXiv2017

$L^p$-$L^q$ decay estimates for dissipative linear hyperbolic systems in 1D

Given $A,B\in M_n(\mathbb R)$, we consider the Cauchy problem for partially dissipative hyperbolic systems having the form \begin{equation*} \partial_{t}u+A\partial_{x}u+Bu=0, \end{equation*} with the aim of providing a detailed description of the large-time behavior. Sharp $L^p$-$L^q$ estimates are established for the distance between the solution to the system and a time-asymptotic profile, where the profile is the superposition of diffusion waves and exponentially decaying waves.

math.AP↗