コーヒーミーティング過去ログ

2026-08-14

What actually happened in the recent OpenAI - Huggingface incident?

シュテッフェン・バッケス (理化学研究所 創発物性科学研究センター (CEMS) 計算物質科学研究チーム 上級研究員 / 理化学研究所 最先端研究プラットフォーム連携 (TRIP) 事業本部 研究DX基盤開発チーム 専門技術員)

In July 2026, OpenAI and Hugging Face disclosed an unusual cybersecurity incident in which an autonomous AI agent, powered by OpenAI models, carried out a multi-stage intrusion into external computer systems. During an internal cybersecurity evaluation, the agent tried to obtain the benchmark solutions directly instead of solving the tasks as intended. It escaped its isolated testing environment through a previously unknown software vulnerability, gained internet access, and eventually infiltrated Hugging Face's production infrastructure, where it performed thousands of automated actions over several days. In this presentation, I will reconstruct how the incident unfolded based on reports from OpenAI and Hugging Face. I will also discuss what makes it different from conventional cyberattacks, what it reveals about the capabilities of autonomous AI agents, and how similar attacks could be detected and defended against in the future.

2026-08-07

What if Quantum Mechanics Were Nonlinear?

松浦 俊司 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 上級研究員)

Linearity is one of the most fundamental principles of quantum mechanics. But what would happen if it broke down? In this talk, I explain this question from the perspective of quantum computation. If quantum mechanics were nonlinear, quantum computers could become extraordinarily powerful, efficiently solving problems believed to be intractable even for ordinary quantum computers. Conversely, the linearity of quantum mechanics imposes fundamental limitations on quantum algorithms for solving nonlinear differential equations,

2026-07-31

Heun Functions as a Common Language: Black Holes Meet Gauge Theory

久保 尚敬 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 基礎科学特別研究員)

Heun functions arise in many problems governed by second-order differential equations with several singular points. I will first give a brief introduction to these functions and explain why they appear in black hole perturbation theory. I will then discuss how quasinormal modes can be formulated as a connection problem for Heun equations, and how gauge theory provides a powerful way to study it. I will conclude with a few examples illustrating how Heun functions appear across various areas of science.

2026-07-24

Singularities are always around/within you

田邊 真郷 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 基礎科学特別研究員)

A singularity of a differentiable map is the location where the differential of the map degenerates. In this talk, I would like to show that this concept appears in many contexts and aspects, focusing on the so-called cusp singularity. I aim to make you (more) familiar with singularities, and also catch a grimple into the vast theory of singularities of differentiable maps.

2026-07-17

Degeneration Meets Berkovich Geometry

後藤 慶太 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 基礎科学特別研究員)

Roughly speaking, a degeneration is a one-parameter family of complex manifolds in which the fibers change their structure drastically at a certain point, say the origin. In this talk, we focus on one of the simplest examples: a degeneration to a node. Through explicit computations, we will observe how the fibers degenerate and how non-Archimedean features emerge in this process. These features naturally lead us to Berkovich geometry.

2026-07-10

God saves quantum computers by playing dice: introduction to quantum error correction

王 啸洋 (理化学研究所 数理創造研究センター (iTHEMS) 数理展開部門 量子数理科学チーム 特別研究員 / 理化学研究所 計算科学研究センター (R-CCS) 連続系場の理論研究チーム 特別研究員)

Microscopic quantum states in quantum computers are extremely sensitive to noise, and their lifetimes are typically limited to only a few microseconds. However, due to no-cloning theorem, we cannot directly identify what kinds of noise have occurred inside a quantum computer and then simply correct the errors led by noise. In this talk, I will explain how quantum error correction overcomes this challenge: it corrects errors without knowing exactly what the errors are. This technique can extend the lifetime of quantum states from a few microseconds to several days, ultimately making quantum computers a powerful key to understanding the microscopic world.

2026-06-19

An Introduction to TQFTs (from a Mathematician’s Viewpoint)

佐野 岳人 (理化学研究所 数理創造研究センター (iTHEMS) 数理展開部門 数学応用研究チーム 研究員)

Topological quantum field theories (TQFTs) are fundamental objects in both physics and mathematics, but their descriptions in the two fields look quite different. In physics, a TQFT is a quantum field theory that is invariant under deformations of spacetime. In mathematics, a TQFT is often formulated as a symmetric monoidal functor from a category of cobordisms to a category of vector spaces. In this talk, I will explain this mathematical formulation and try to bridge the gap between the mathematical definition and the physical intuition. I will also briefly describe how TQFTs appear in mathematics, in particular in the construction of knot homology theories.

2026-06-12

But how many baskets should you put your eggs in?

ブライアン・アンドリュー・ミンツ (理化学研究所 数理創造研究センター (iTHEMS) 数理展開部門 数理社会科学チーム 特別研究員)

“Don’t put all your eggs in one basket” is old advice against investing too much in any single area, but it does not tell us how many baskets to use, or how many eggs to place in each one. This everyday question contains a deep mathematical problem: how should we choose between different ways of distributing risk? Depending on the precise formulation, the “best” strategy can change completely; or in some cases, not exist at all. We will trace the history of answers to this simple yet profound question, connecting the work of Nobel prize winners across a variety of disciplines including biology, mathematics and economics. Understanding this topic even made one mathematician a multi-millionaire!

2026-06-05

Spontaneous symmetry breaking

本多 正純 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 上級研究員 / 埼玉大学 大学院理工学研究科 連携教授)

Why is iron strongly attracted to magnets? Why do elementary particles such as electrons have mass? The answers to these questions are closely related to a common idea called spontaneous symmetry breaking. In my talk, I will introduce the idea of spontaneous symmetry breaking starting with a demonstration experiment. I am also going to mention speculative examples of spontaneous symmetry breaking in various areas beyond physics.

2026-05-29

Current "virus trends": hanta, ebola, et cetera

カトゥリン・ボシゥメン (理化学研究所 数理創造研究センター (iTHEMS) 副センター長 / 理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 部門長代理 / Professor, Department of Physics, Toronto Metropolitan University, Canada)

I will discuss some viruses that have been in the news lately, what they are, where they come from, and explain whether and to what extent you should be worried about them. I'll also tell you about some cool and weird stuff I found out as part of preparing this short talk.

2026-05-22

The Origami Classification of Cosmic Structures

デリック・ビーティー・インマン (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 研究員)

The cosmic web forms from the gravitational collapse of a nearly uniform density field. I will discuss how this process proceeds starting from the first light objects to the present massive ones. Along the way, I will show an analogy between phase-space folding and origami which can be used to classify different types of structures.

2026-05-15

Exploring the Minimal Scale and Complexity of Networks that Generate Intelligence

風間 北斗 (理化学研究所 脳神経科学研究センター (CBS) 知覚神経回路機構研究チーム チームディレクター)

When individual elements interact within a system, there exists a transition point where qualitatively new phenomena arise—properties that cannot be reduced to a single component. For example, in inorganic systems, this transition can manifest as a phase transition, while in organic systems, it can give rise to life-specific functions. However, a fundamental question remains: what is the minimum cellular scale and network complexity required to generate higher brain functions? Moving beyond anthropocentric viewpoints, we will investigate the structural and functional principles of intelligence within the broader context of life. By cross-examining gene regulatory networks, neural connectivity, and functional dynamics, we aim to uncover the essential architectural motifs that define the origins of biological intelligence.

2026-04-24

GW ringdowns and echoes

チュユ・チェン (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 基礎科学特別研究員)

Typical gravitational wave (GW) waveforms emitted by binary black hole mergers contain three phases, including inspiral, merger, and ringdown. Beyond the standard picture of classical black holes, there could be series of GW echoes following the standard ringdown signals, whose existence could be an indicator of horizon-scale new physics. In this talk, I will briefly introduce the physics of ringdown and GW echoes.

2026-04-10

The transition to synchronization: from phase reduction to the Kuramoto model

リッカルド・ムオロ (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 基礎科学特別研究員)

Synchronization is one of the most studied phenomena in the field of complex systems and a paradigmatic example of self-organizing behavior. The synchronization of coupled oscillators was first observed by Christiaan Huygens, who in the 17th century noticed that pendula hanging on the same wall tend to synchronize in anti-phase. Later research highlighted the occurrence and relevance of synchronization in many natural and artificial systems, ranging from fireflies’ blinking and frogs’ croaking to rhythmic contractions in cardiac cells and neuronal activity, and from bridge oscillations to power angles in electrical power grids. The key factor enabling this collective behavior is the interaction among oscillators, which can be, for example, mechanical (bridges, pendulum clocks), electrical (heart, power grids), visual (fireflies), or acoustic (frogs). Until about sixty years ago, it was unclear how synchronization emerges. A major step forward came from Winfree, who proposed describing oscillators using only their phases and conjectured that a transition to synchronization would occur above a critical coupling strength. This idea inspired Kuramoto, who introduced his celebrated model and provided a solution to this longstanding problem. In this short talk, I will introduce the phase reduction approach and retrace the path that led Winfree and Kuramoto to these results. I will then discuss the Kuramoto model and illustrate the transition to synchronization.

2026-03-27

Condensation Tensor Network

大森 寛太郎 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 上級研究員)

A matrix is an array of numbers with two indices, composed along a line. A tensor has more indices, and tensors can be composed along a network that fills a region of space — much like how data, signals, or physical states are organized spatially in many areas of science. What conditions on the individual tensors guarantee that the network as a whole has nice properties? I will introduce the notion of a condensation tensor network, where simple algebraic conditions on each tensor — checkable locally — ensure that the entire network behaves coherently at any scale. The resulting structure defines a projection onto a subspace that is independent of the system size. These networks provide a unified language for topological phases of matter and quantum error correction.

2026-03-13

Dipole symmetry from anomaly

戎 弘実 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 研究員)

Anomalies often signal that a quantum system cannot realize a trivial symmetric phase. In lattice systems, certain combinations of internal and spatial symmetries lead to such anomaly constraints. We discuss that dipole symmetry, a spatially modulated symmetry, can naturally emerge from these anomaly structures. Starting from anomalous lattice models, we demonstrate that gauging part of the internal symmetry leads to the appearance of dipole symmetry in the resulting theory. This perspective provides a simple way to understand the origin of modulated symmetries and highlights how anomaly considerations can give rise to unconventional symmetry structures in quantum lattice systems.

2026-03-06

How many sequences of games are possible in Koshien?

孔 星植 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 研究員)

I address the question “How many sequences of games are possible in Koshien?” using combinatorial techniques from mathematical phylogenetics. I model the tournament as a labeled evolutionary history and apply tree enumeration methods to count the number of distinct outcome sequences. By interpreting each game as an internal node in a rooted labeled history, the problem reduces to counting labeled histories consistent with the tournament structure. I start the talk with basic introduction of evolutionary trees and try to establish a connection between single-elimination tournaments and the combinatorics of evolutionary trees.

2026-02-27

What Are Modular Forms?

村上 友哉 (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 研究員)

In this talk, I introduce the concept of modular forms and explain some of their surprising applications to number theory, topology, and mathematical physics. Modular forms are holomorphic functions characterized by rich symmetry properties. At first glance, their definition appears purely analytic.Nevertheless, they contain deep arithmetic information and play a central role in modern number theory. In recent developments, modular forms have also emerged in the context of topological quantum field theory. This talk aims to provide an accessible overview of these ideas for a general audience.

2026-02-20

How a player can unilaterally control payoffs in repeated games

村瀬 洋介 (理化学研究所 数理創造研究センター (iTHEMS) 数理展開部門 数理社会科学チーム チームディレクター / 理化学研究所 計算科学研究センター (R-CCS) 離散事象シミュレーション研究チーム 上級研究員)

In repeated interactions, players can choose their actions based on past outcomes, leading to a wide variety of possible strategies. In this talk, I introduce "zero-determinant" strategies, discovered by the famous physicist Freeman Dyson, which show that in repeated games a player can unilaterally enforce linear relations between long-run average payoffs, independent of the opponent's strategy. I will explain the basic idea using a Markov-chain formulation of the repeated Prisoner’s Dilemma.

2026-02-13

Failure of Scale Separation in Physics

ウェイシャン・シャオ (理化学研究所 数理創造研究センター (iTHEMS) 数理基礎部門 基礎科学特別研究員)

The success of modern physics relies heavily on the principle of scale separation, which allows complex systems to be described by simplified effective theories that are largely insensitive to microscopic details. In this talk, I will review this Wilsonian framework and illustrate its usefulness across a wide range of physical contexts. I will then discuss situations in which this separation of scales breaks down, focusing on examples motivated by quantum gravity as well as by certain physical settings in our universe. These examples show how high-energy physics can have unexpectedly strong effects at large scales.