Seminar

Optimal transport: Thermodynamics, biological information processing and generative models

January 18 (Mon) 13:30 - 15:00, 2027

Sosuke Ito (Associate Professor, Graduate School of Science, The University of Tokyo)

This is a joint seminar co-hosted by Institute for Physics of Intelligence (iπ), UTokyo, and the Venue is #1042,10F of Science Bldg.1 Optimal transport theory is a useful theoretical framework for geometrically characterizing the distance between probability distributions. It is not only used in recent research on generative models but has also been applied to detect information-processing capabilities in biological systems from a statistical perspective, and to discuss thermodynamic efficiency. In this seminar, I will begin with an introduction to optimal transport theory, and also discuss our research on its various applications. More specifically, I will discuss the following topics: 1) The problem of minimizing entropy production within a finite time and its experimental testing. 2) The application of optimal transport to the training of generative models, specifically diffusion models, and their optimal protocols. 3) Quantifying information processing performance using optimal transport in biological signal transduction.

Venue: Faculty of Science Bldg.1, School of Science, The University of Tokyo, Hongo Campus (Main Venue) / via Zoom

Event Official Language: English

Hamiltonian simulation: Standard algorithms and recent developments

December 17 (Thu) 10:30 - 17:00, 2026

Kaoru Mizuta (Associate Professor, Center for Quantum Information and Quantum Biology, The University of Osaka)

This is a lecture-style seminar of quantum computation. The tentative schedule consists of three slots: 10:30-12:00, 13:30-15:00, and 15:30-17:00. Due to equipment limitations, the blackboard might not be visible on Zoom, so we recommend attending in person.

Venue: Seminar Room #359 (Main Venue) / via Zoom

Event Official Language: English

Chiral Anomaly, Neutral-Pion Lifetime and Axion–Photon Coupling

December 7 (Mon) 15:00 - 16:30, 2026

Xu Feng (Professor, Director, Institute of Theoretical Physics, School of Physics, Peking University, China)

The chiral anomaly provides one of the rare opportunities to test QCD in the confinement regime at the percent level and beyond. A prime example is the neutral pion lifetime: after two decades of PrimEx measurements reaching 1.5% precision, future experiments at an upgraded 22 GeV JLab aim at the sub-percent level, calling for a comparably precise first-principles QCD prediction. The same anomaly also governs the QCD contribution to the axion–photon coupling, a key quantity connecting axion models to experimental searches for one of the leading dark-matter candidates. In this talk, I will present our recent lattice QCD studies of these two quantities, going beyond the chiral limit.

Venue: Hybrid Format (3F #359 and Zoom), Seminar Room #359 (Main Venue) / via Zoom

Event Official Language: English

Quantum Geometry from Matrices: Framework and Applications

November 30 (Mon) 15:00 - 16:30, 2026

Harold Steinacker (Senior Scientist, Faculty of Physics, University of Vienna, Austria)

I will discuss the general framework of matrix geometry, which is a version of quantum geometry defined by a set of almost-commuting matrices. This framework allows for an efficient description of manifolds embedded in some target or feature space, which is particularly efficient for high-dimensional manifolds. I will discuss some applications in data science as well as in fundamental physics, notably how spacetime and gravity can emerge from the IKKT or IIB matrix model in this framework.

Venue: Hybrid Format (3F #359 and Zoom), Main Research Building / via Zoom

Event Official Language: English

Rare k-mers reveal centromere haplogroups underlying human diversity and cancer translocation

November 26 (Thu) 13:00 - 14:00, 2026

Yuichi Shiraishi (Chief, Division of Genome Analysis Platform Development, National Cancer Center Japan)

Centromeres are highly variable and rapidly evolving regions of the human genome, and are frequently implicated in various cancers. Human centromeres consist of highly repetitive DNA sequences known as α-satellite higher-order repeats (HORs), which makes their analysis using conventional genome sequencing approaches extremely challenging. Although long-read sequencing technologies now enable near-complete sequencing of centromeres, applying these methods to large numbers of samples remains difficult. As a result, our understanding of centromere diversity across populations and its role in chromosomal rearrangements in cancer has remained limited. In this study, by focusing on rare k-mer markers, we demonstrate that α-satellite HOR haplotypes cluster into genetically related haplogroups. We further developed a software tool, ascairn, to accurately infer these haplogroups from common short-read whole-genome sequencing (WGS) data. Applying our approach to a dataset of more than 3,300 human samples revealed extensive centromere diversity and geographic variation. We also applied the rare k-mer–based approach to centromere sequences from 104 Japanese individuals generated as part of the Japanese Pangenome Project. This analysis revealed characteristic patterns of centromere evolution in the Japanese population and provided new insights into Japanese-specific centromere variation. In addition, we used this method to investigate class-defining chromosomal translocations, specifically der(1;19), also known as 1p/19q co-deletion, in oligodendrogliomas. By analyzing 142 samples from TCGA, we hypothesized that the breakpoints occurred on the p-arm side of D1Z7 and the q-arm side of D19Z3. This hypothesis was subsequently validated through extensive long-read sequencing in two cases, indicating that der(1;19) results in a dicentric chromosome containing two functional centromeres. We also observed that chromosomal rearrangements preferentially occurred in haplotypes with longer repetitive arrays on either the q-arm or p-arm side of the centromere. Furthermore, we identified specific centromere haplogroups associated with either increased or decreased likelihoods of der(1;19) translocations. Our findings show that rare k-mer analysis is a powerful approach for investigating centromere complexity and clarifying the mechanisms underlying centromere-associated abnormalities in cancer.

Venue: via Zoom / Seminar Room #359

Event Official Language: English

Quantum Simulation of Non-Equilibrium Dynamics in SU(2) and SU(3) Lattice Gauge Theories

November 24 (Tue) 15:00 - 16:30, 2026

Xiaojun Yao (Research Assistant Professor, University of Washington, USA)

Venue: Seminar Room #359 (Main Venue) / via Zoom

Event Official Language: English

Real-time calculations for excited state tunneling

November 17 (Tue) 14:00 - 15:30, 2026

Thomas Steingasser (Postdoctoral Fellow, Institute of Theoretical Physics (IFT), Spain)

Quantum tunnelling is conventionally formulated through Euclidean instantons, obscuring the role of physical time and complicating non-vacuum initial states. I will review the steadyon picture, in which infinitesimally regulated real-time path integrals are controlled by complex saddles whose projections recover familiar instantons. I will summarize its applications to finite-temperature decay, resonance states and states carrying conserved Noether charge, highlighting an emerging unified description of tunnelling beyond the false vacuum.

Venue: Hybrid Format (3F #359 and Zoom), Seminar Room #359 / via Zoom

Event Official Language: English

A non-negative matrix factorization-based approach to integrating multimodal omics data with automatically tuned hyperparameters

November 16 (Mon) 14:00 - 15:00, 2026

Dorothy Ellis (Postdoctoral Researcher, Laboratory for Integrative Genomics, RIKEN Center for Integrative Medical Sciences (IMS))

Venue: via Zoom

Event Official Language: English

Lorentzian geometry beyond smoothness and the space of Cauchy hypersurfaces

November 13 (Fri) 14:00 - 15:15, 2026

Keita Takahashi (Postdoctoral Researcher, Mathematical Application Research Team, Division of Applied Mathematical Science, RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS))

Lorentzian geometry is the study of manifolds equipped with a metric of signature $(-,+,\dots,+)$. It provides the mathematical foundation of general relativity, in which gravity is described by the curvature of spacetime. As exemplified by the singularity theorems of Hawking and Penrose in the 1960s, the field has developed through a close interplay between mathematics and physics. Although Lorentzian geometry looks formally similar to Riemannian geometry, it behaves very differently. The Lorentzian metric does not induce a distance in the usual metric-space sense. Instead, the time separation satisfies a reverse triangle inequality for causally ordered points, and causal relations play a central role in the global geometry of spacetime. In recent years, ideas from metric geometry, in particular synthetic theories of curvature bounds such as Alexandrov spaces and RCD spaces, have been brought into Lorentzian geometry. This has given rise to the rapidly developing theory of Lorentzian length spaces and their synthetic curvature bounds. In this talk, I will introduce Lorentzian geometry for non-specialists, emphasizing the differences from Riemannian geometry and surveying recent developments. I will then discuss my work on the space of Cauchy hypersurfaces of a globally hyperbolic spacetime, viewed as a metric space in its own right.

Venue: Hybrid Format (3F #359 and Zoom), Main Research Building / via Zoom

Event Official Language: English

122nd Data Assimilation and Prediction Science Seminar

November 12 (Thu) 15:30 - 17:30, 2026

Filipe Aires (Research Director, LERMA, CNRS, France)

Title: Thirty Years of Artificial Intelligence in Satellite Remote Sensing: From Atmosphere to Surface Waters Abstract: For over 30 years, our research has explored the potential of machine learning and artificial intelligence for satellite remote sensing of the Earth system. This presentation will highlight two major research directions. First, we will discuss atmospheric remote sensing using the IASI instrument, focusing on the retrieval of atmospheric temperature and humidity profiles, as well as cloud characterization. Second, we will present our activities in monitoring continental surface waters, wetlands, and peatlands, combining multi-satellite observations and advanced AI methodologies, with particular emphasis on recent developments enabled by the SWOT mission. These applications illustrate the growing contribution of AI to Earth observation.

Venue: Hybrid Format (RIKEN R-CCS room C104-2 and Zoom) / via Zoom

Event Official Language: English

Topological properties of empirical fitness landscapes from merge trees

November 9 (Mon) 13:00 - 14:00, 2026

Igors Dubanevics (Ph.D. Student, Okinawa Institute of Science and Technology Graduate University (OIST))

How evolution navigates sequence space depends on the shape of the fitness landscape. Viewed as a scalar field over a discrete genotype graph, theoretical landscapes range from single-peaked to rugged and fragmented, with very different mutational accessibility. Which shapes real biological sequences occupy remains an open empirical question. Deep mutational scanning now measures sequence–function maps across complete combinatorial regions of sequence space, but the resulting landscapes are large, noisy, and hard to compare across studies. Existing analyses typically reduce each dataset to a few summary statistics, discarding which peaks connect to which, and through which saddles. We are developing a two-layer toolkit for empirical sequence–function landscapes. A statistical layer models overdispersed sequencing counts directly to separate genuine phenotypic differences from measurement noise, pruning spurious topological features before analysis. A topological layer then summarises the landscape as a merge tree of peaks, saddles, and the high-fitness ridges that link them, using persistence as a natural scale to distinguish global structure from minor ruggedness. I will first illustrate this approach on the largest published combinatorially complete protein-coding landscape, spanning over 260,000 sequences. Filtering by measurement uncertainty and persistence simplifies an apparently rugged surface, revealing that major peaks remain connected through high-fitness saddles rather than isolated by deep valleys. I will then present preliminary findings from a comparative atlas of more than 150 combinatorially complete sequence–function landscapes spanning DNA, RNA, and protein. By comparing merge-tree structure across this collection, we ask whether high-fitness ridges that bypass rugged regions recur across unrelated sequences, and whether landscapes group by the biological function they encode.

Venue: via Zoom / Hybrid Format (3F #359 and Zoom), Seminar Room #359

Event Official Language: English