Home
On Mon September 07, 2026

Speaker

Sang Won Bae

Title

Counting Polygons and Holes: Algorithms and Complexity

Abstract

Given a finite set S of points in the plane in general position, a polygon of k corners chosen from S is called a k-gon of S, and a k-gon is called a k-hole if it contains no other points of S in its interior. In this talk, we are interested in the very fundamental problem of counting k-gons or k-holes of S for a given set S and an integer k. Considering the computational complexity of these counting problems, this talk covers known combinatorial results on k-gons and k-holes, algorithms that count them, lower bounds, and open research problems from the theoretical point of view.

Bio

Sang Won Bae received a PhD in computer science at KAIST in 2008. His research interests include research problems in discrete and computational geometry and their applications. Currently, he works as a professor in Kyonggi University.

Language

Korean (Offline)

On Mon September 14, 2026

Speaker

Moontae Lee

Title

Self-Improving AI Agents

Abstract

TBD

Bio

TBD

Language

English (Offline)

On Mon September 21, 2026

Speaker

YongKeun (Paul) Park

Title

Holotomography and AI

Abstract

Holotomography (HT) provides label-free, quantitative three-dimensional (3D) imaging of live cells and organoids by mapping their intrinsic refractive index (RI) distributions [1,2]. In analogy to X-ray computed tomography, HT records multiple holograms of a sample under varied illumination angles and reconstructs the 3D RI distribution by inverting the wave equation [1,6]. The resulting data are inherently quantitative and free of exogenous labels, yet they are also physically abstract - RI is not a molecular identity - and each measurement yields gigabyte-scale volumes. Both properties make HT an unusually good substrate for machine learning: the physics fixes the units, and the learning supplies the interpretation.
In this presentation, I will discuss what the combination of HT and artificial intelligence (AI) has delivered so far and where it still breaks down, across three tasks: segmentation of unlabeled subcellular structures, classification of cell types and states [4], and inference of molecular contrast - including virtual staining of fluorescence-like or histology-like images from RI volumes [5,7,8]. For each, I will address the limitations that matter in practice: the scarcity of paired ground-truth data, the difficulty of validating an inference that has no label to compare against, generalization across instruments and sample-preparation protocols, and the interpretability required for clinical adoption. I will close with the directions I consider most consequential - physics-informed reconstruction [3,9], standardized RI measurement across platforms [6], and prospective clinical validation - and with what would be needed for label-free 3D imaging with AI to become a routine tool in biology and diagnostics.

References:
1. Y. Park, C. Depeursinge and G. Popescu, Nature Photonics 12 (10), 578-589 (2018)
2. Y. Baek and Y. Park, Nature Photonics 15 (5), 354-360 (2021)
3. S. Shin and Y. Park, Nature Materials 21, 317-324 (2022)
4. Y. Jo et al., Nature Cell Biology 23, 1329-1337 (2021)
5. J. Park et al., Nature Methods 20, 1645-1660 (2023)
6. G. Kim et al., Nature Reviews Methods Primers 4, 51 (2024)
7. C. Oh et al., Nature Communications 16, 1685 (2025)
8. J. Park et al., Nature Communications 16, 4781 (2025)
9. J. Lee et al., Nature Photonics 21, 1 (2026)

Bio

YongKeun (Paul) Park is an Endowed Chair Professor of Physics at KAIST and Co-founder and CEO of Tomocube. His research focuses on holotomography, quantitative phase imaging, and artificial intelligence for label-free 3D biomedical imaging. He has pioneered applications of these technologies in cell biology, organoids, regenerative medicine, and clinical diagnostics. He is a Fellow of Optica and SPIE.

Language

English (Offline)

On Mon September 28, 2026

Speaker

Suthee Ruangwises

Title

Card-Based Cryptography: Computing Securely with Playing Cards

Abstract

Can we perform cryptography using only a deck of playing cards? By representing private information with face-down cards and manipulating them through simple operations such as shuffling, it is possible to perform various cryptographic tasks while keeping it hidden. In this talk, we will introduce the history and basic concepts of card-based cryptography and give an overview of its three main lines of research: secure multi-party computation, where participants jointly compute a function without revealing their private inputs; zero-knowledge proofs, where a prover demonstrates that a statement is true without revealing any additional information; and player simulation, where cards are used to simulate the private actions of a player in games such as UNO. Along the way, we will demonstrate simple protocols from each of these areas and highlight some recent developments in the field.

Bio

Suthee Ruangwises is currently a lecturer at Chulalongkorn University, Thailand. Suthee's research interests include card-based cryptography and the computational complexity of puzzles. Suthee received the B.S. degree from Massachusetts Institute of Technology in 2015, the M.S. and Ph.D. degrees from Tokyo Institute of Technology in 2017 and 2020, respectively, and was previously a postdoctoral researcher at Tokyo Institute of Technology and The University of Electro-Communications, Japan.

Language

English (Offline)

On Mon October 12, 2026

Speaker

Johannes Blömer

Title

Sums of Square Roots, Travelling Salesmen and Diophantine Approximation

Abstract

In geometric problems where the inputs are points in Euclidean space — for example, the Euclidean travelling salesman problem — comparing the quality of different solutions comes down to determining the sign of a sum of square roots. However, it is not known how to do this in polynomial time since the best-known lower bounds for the distance of a sum of square roots to zero are doubly exponential in the number of summands. Consequently, it is not known whether the Euclidean travelling salesman problem (ETSP) is in NP. This talk will discuss the current state of research on the sums of square roots problem. In particular, we will discuss the close connection between the sums of square roots problem and quantitative versions of the celebrated Thue–Siegel–Roth theorem and the subspace theorem from diophantine approximation. The talk is partially based on joint work with Friedrich Eisenbrand.

Bio

Since 2000 Prof. Dr. Johannes Blömer is a professor in the Department of Computer Science of Paderborn University, where he leads the Codes and Cryptography group. He has done work in various areas of theoretical computer science, including complexity theory, clustering, algorithmic number theory and, (post-quantum) cryptography and security. His work in cryptography ranges from number theoretic foundations, e.g. in lattice theory, to the construction and formal analysis of cryptographic primitives and to practical aspects, e.g. side-channel attacks and secure implementations. His work has been published in leading conferences and journals of computer sciences and cryptography like CRYPTO, Eurocrypt, ACM CCS, ICALP, FOCS, and STOC. He served on various program committees and is on the board of two journals in theoretical computer science.

Language

English (Offline)

On Mon October 26, 2026

Speaker

Naoto Yokoya

Title

Open and Globally Reliable GeoAI

Abstract

This lecture presents our research toward open and globally reliable GeoAI that understands, reasons about, and reconstructs the real world from diverse and incomplete observations. Topics include global semantic mapping, multimodal disaster understanding, 3D reconstruction and scene generation from images and language, spatial reasoning with vision-language models, and confidence calibration for VLMs and AI agents. Through these examples, I will discuss how visual and spatial intelligence can become more generalizable, accessible, and trustworthy.

Bio

He is a Professor at The University of Tokyo and the Team Director of the Geoinformatics Team at the RIKEN Center for Advanced Intelligence Project. His research interests include remote sensing, computer vision, and machine learning, with applications to disaster management and environmental assessment. He has been recognized as a Clarivate Highly Cited Researcher every year since 2022.

Language

English (Offline)

On Mon November 02, 2026

Speaker

Seungbum Jo

Title

Encoding Data Structures on Arrays

Abstract

Queries on arrays are fundamental in computer science and arise in areas such as databases, computational biology, and geographic information systems. The goal is to answer queries about positions or intervals quickly while keeping the space usage small. In this talk, I will survey encoding data structures for queries on arrays, including both range queries and position-based queries. These structures store only the information needed to answer the queries, without access to the original array. I will present the main ideas and representative results on space-efficient encodings for several important array query problems.

Bio

Seungbum Jo is an Assistant Professor in the Department of Computer Science at Chungnam National University. He received his B.S. and M.S. from KAIST and his Ph.D. from Seoul National University. Before joining Chungnam National University, he held postdoctoral positions at Université libre de Bruxelles, the University of Haifa, and the University of Siegen. His research interests include algorithms and data structures, especially space-efficient data structures for arrays and graphs, sorting, and algorithm engineering.

Language

English (Offline)

On Mon November 09, 2026

Speaker

Eunjin Oh

Title

How Geometry Changes the Computational Landscape

Abstract

Many fundamental graph problems are computationally hard on arbitrary inputs, but their complexity can change dramatically when the input carries geometric structure. This talk asks two questions: What geometric structures can graphs have, and why do those structures lead to faster algorithms? To address the first question, I will begin by examining three ways in which graphs can encode geometry: through crossing-free embeddings, through shortest-path distances that approximate Euclidean distances, and through intersections among geometric objects. These representations define distinct graph classes, and even small changes in the underlying geometric assumptions can lead to sharp changes in computational complexity. To address the second question, I will show that geometry can make algorithms faster by providing structures unavailable in general graphs. Clique-based separators decompose unit disk graphs into smaller subproblems, while compact representations avoid explicitly processing the quadratic-size graph underlying geometric multimatching. Taken together, these examples illustrate how geometry changes the computational landscape.

Bio

Eunjin Oh is an associate professor in the Department of Computer Science and Engineering at POSTECH. Her research bridges graph algorithms and computational geometry, with a focus on geometric optimization, shortest-path and distance problems in geometric graphs, and parameterized and approximation algorithms. She received her B.S. and Ph.D. in Computer Science and Engineering from POSTECH, completing her Ph.D. in 2018. After a postdoctoral fellowship at the Max Planck Institute for Informatics in Germany, she joined the POSTECH faculty in 2019 and became an associate professor in 2023. She currently serves as an editor for Theoretical Computer Science and the Journal of Computational Geometry and served as a program co-chair for WADS 2025.

Language

English (Offline)

On Mon November 16, 2026

Speaker

Donghee Na

Title

AI는 소프트웨어 엔지니어링을 어떻게 바꾸고 있는가

Abstract

AI 시대를 맞이하면서, 2010년대에 우리가 익숙하게 받아들였던 소프트웨어 엔지니어의 정의가 빠르게 변화하고 있습니다. 이번 강좌에서는 AI가 소프트웨어 개발 방식과 엔지니어의 역할을 어떻게 바꾸고 있는지 살펴보고, 그 과정에서 우리가 새롭게 마주하게 된 문제와 고민에 대해 함께 이야기해보고자 합니다.

Bio

TBD

Language

Korean (Offline)

On Mon November 23, 2026

Speaker

Byungseok Roh

Title

From Pretraining to Production: Building AI Models for Millions of Users

Abstract

TBD

Bio

Byungseok Roh is the Head of Unified Foundation Model Studio and an Executive Leader at Kakao, where he leads the development of large-scale foundation models spanning language, vision, and audio. His organization builds AI models end-to-end, from pretraining LLMs to Vision-Language Models (VLMs), Audio Language Models, and omni-modal language models, while also focusing on large-scale training, inference optimization, and production deployment. Previously, he served as a Research Team Lead at Kakao Brain, a Technical Lead at Intel, and a Researcher at ETRI.

Language

Korean (Offline)

On Mon November 30, 2026

Speaker

Sangdoo Yun

Title

Beyond Scaling: Efficient and Grounded Multimodal Intelligence

Abstract

Scaling has enabled multimodal models with impressive perceptual and linguistic capabilities, but real-world intelligence requires more than scale alone. In this talk, I will present my research through a unifying question: what makes a useful representation? I will trace this question through efficient adaptation in weight space, selective memory through KV-cache compression, and grounded world modeling and robotics. Across these stages, I explore how representations can remain compact, adaptable, and grounded. I will conclude with a research agenda for efficient and grounded multimodal intelligence, centered on where knowledge should reside (in features, weights, memory, or the external world) and how it can be used under real-world constraints.

Bio

Sangdoo Yun leads NAVER’s Future AI Center and NAVER AI Lab and is an adjunct professor at the Artificial Intelligence Institute of Seoul National University. His research focuses on building efficient, reliable, and capable multimodal foundation models for real-world AI applications. At NAVER, he has led the development of machine learning methods spanning model architecture, training and optimization, robustness, multimodal learning, and large-scale AI systems. He is best known for co-inventing CutMix and has contributed to advances in model architecture, optimization, robustness, and multimodal learning. He received his B.S., M.S., and Ph.D. degrees from Seoul National University in 2010, 2013, and 2017, respectively, with his graduate research focused on computer vision. He regularly serves as an Area Chair for leading AI conferences, including NeurIPS, CVPR, ICLR, ECCV, ICCV, and ICML.

Language

Korean (Offline)

On Mon December 07, 2026

Speaker

Maximilian Gorsky

Title

How Graph Structure Makes Hard Problems Tractable

Abstract

A central aim of theoretical computer science and of algorithms research in particular is to understand what makes computational problems difficult. In this talk, I will introduce parameterized algorithms, an area that asks a deceptively simple question: What structure in an input can make an otherwise intractable problem efficiently solvable? We will explore this question through a fundamental problem in graph theory: Can specified pairs of points in a network be connected by paths that do not intersect? This problem and its variants arise naturally in applications such as chip design and also play a central role in the development of many other graph algorithms. Using this problem as an example, I will introduce several powerful ideas from modern algorithms research, including advanced forms of dynamic programming, techniques for reducing an instance to its essential core, and structural approaches that bring order to complicated graphs. Our example illustrates how graphs became such an important testing ground for new algorithmic techniques over the last decades. More broadly, the talk will offer an accessible introduction to how researchers in theoretical computer science think: how they identify the source of a problem’s difficulty, uncover useful structure, and turn that structure into algorithms.

Bio

Maximilian Gorsky is a computer scientist working in the intersection of efficient algorithms and structural graph theory. Having received his doctorate at the TU Berlin in 2024 under the supervision of Stephan Kreutzer, he has been working at the Institute for Basic Science in the Discrete Mathematics Group of Sang-il Oum since November of 2024. In 2025 he was named an IBS Researcher of the Year.

Language

English (Offline)