2023 Fall Semester
[2023 Fall Semester]
지정 날짜 목요일 12:00
운영위원 : 권혜민 ( khmin1121@ajou.ac.kr, 팔달관 622호 )
감독위원 : 표성인( luinn27@ajou.ac.kr, 팔달관 432호 )
1. 9/14 (발표자 : 강미연-데이터사이언스전공/박사과정)
Title : Introduction to persistent homology
Abstract : Persistent homology, a powerful mathematical tool in the field of topological data analysis, is utilized to conduct multi-scale analyses on a given set of data points. This method aids in the identification and characterization of clusters, voids, and holes within the dataset. By capturing topological features at different scales, persistent homology provides valuable insights into the underlying structures of complex datasets, making it a fundamental tool in modern data analysis and visualization.
2. 10/05 (발표자 : 윤영한-수학전공/박사수료)
Title : The cohomology rings of real permutohedral varieties
Abstract : In this talk, we establish explicit descriptions of the cohomology ring of real permutohedral varieties. In particular, the multiplicative structure is given in terms of alternating permutations.
3. 10/12 (발표자 : 권혜민-수학전공/박사수료)
Title : Odd coloring and strong odd coloring
Abstract : Petr\v sevski and \v Skrekovski introduced an odd coloring of a graph $G$, which is a relaxation of a proper coloring of the square of $G$. An odd $k$-coloring is a proper $k$-coloring such that every non-isolated vertex has a color appearing an odd number of times on its neighborhood. Naturally, we could obtain a strong version of an odd coloring, which is a strong odd coloring: a strong odd $k$-coloring is a proper $k$-coloring such that for every non-isolated vertex, each color on its neighborhood appears an odd number of times. The minimum $k$ for which $G$ has a strong odd $k$-coloring is the strong odd chromatic number of $G$, denoted $\chi_{so}(G)$. We present results on $\chi_{so}(G)$ for a sparse graph $G$ and compare them with the results of an odd coloring of $G$ and a proper coloring of the square of $G$. This talk is based on joint work with Eun-Kyung Cho, Ilkyoo Choi, and Boram Park.
4. 11/02 (발표자 : 서승원-데이터사이언스전공/석사과정)
Title : Shifting perspectives: from Epilepsy to cerebroelectric disorder
Abstract : We performed LDA (Latent Dirichlet Allocation) topic modeling on words using the text data analysis algorithm on NAVER news article crawling text data by year. To explain how words change over time for each year, the “Dynamic Word Embedding” method can be used to increase explanatory power.
5. 11/09 (발표자 : 이규철-데이터사이언스전공/석사과정)
Title : Convolutional Neural Network models for image classification in the ImageNet Challenge
Abstract : Convolutional Neural Network (CNN) models have achieved groundbreaking advancements in the field of image classification. Particularly, in large-scale image recognition competitions like the ImageNet Challenge, CNN models have demonstrated outstanding performance. This presentation provides a brief introduction to CNNs and reviews and compares key CNN models used for image classification in the ImageNet Challenge, analyzing the strengths and weaknesses of each model.
6. 11/16 (발표자 : 장설-데이터사이언스전공/석사과정)
Title : Defense-Gan: Protecting classifiers against adversarial attacks using generative models
Abstract : In recent years, deep neural network approaches have been widely adopted for machine learning tasks, including classification. However, they were shown to be vulnerable to adversarial perturbations: carefully crafted small perturbations can cause misclassification of legitimate images. We propose Defense-GAN, a new framework leveraging the expressive capability of generative models to defend deep neural networks against such attacks. Defense-GAN is trained to model the distribution of unperturbed images. At inference time, it finds a close output to a given image which does not contain the adversarial changes. This output is then fed to the classifier. Our proposed method can be used with any classification model and does not modify the classifier structure or training procedure. It can also be used as a defense against any attack as it does not assume knowledge of the process for generating the adversarial examples. We empirically show that Defense-GAN is consistently effective against different attack methods and improves on existing defense strategies.
7. 11/23 (발표자 : 장현태-수학전공/박사수료)
Title : On the nonsingular complete toric varieties of picard number $4$
Abstract : A non-singular complete fan can be characterized by the underlying simplicial complex $K$ of the fan and the set of primitive ray vectors of $1$-dimensional cones. In particular, $K$ is a PL-sphere, and the set of the ray vectors gives a map $\lambda$ from the vertices to the set of integer vectors such that the image of each face under the map is a unimodular set. Such $\lambda$ is called a characteristic map over $K$. It is known that PL-spheres admitting a characteristic map are completely listed by the way of wedge operation, and there is a complete list up to Picard number $4$. In this talk, we provide how this classification can be used to test a theory (or a hypothesis) in toric manifolds by an example associated to minimal components in the normalized space of rational curves on a toric manifolds. This is joint work with Suyoung Choi and Mathieu Vall\'ee.
8. 11/30 (발표자 : 김민규-수학전공/박사수료)
Title : The balanced property of subshifts
Abstract : In this talk, we introduce balanced property of subshitfs. This property is related to special measures, called Gibbs measures, on subshifts and also related to the almost specification property of subshifts.
9. 12/07 (발표자 : 정경진-데이터사이언스전공/석사과정)
Title : Object Detection for Fish Species Identification and Classification
Abstract : 본 세미나는 [2023 제3회 K-water AI 경진대회]에서의 어종(魚種) 식별 및 분류 알고리즘 개발 과정과 결과에 대한 소개입니다. 세미나는 Object Detection 기술의 핵심 개념을 간략히 소개한 후, 대회에서 적용한 YOLO (You Only Look Once) 모델에 중점을 두어 세부적으로 설명합니다.