DEEP-IN Seminar
42 events
-
SeminarOptimal 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
-
Seminar
Multi-Agent Scientific Discovery in Theoretical Quantum Physics
October 14 (Wed) 15:30 - 16:30, 2026
Sirui Lu (Ph.D. Candidate, Max Planck Institute of Quantum Optics and TU Munich, Germany)
This is a joint seminar co-hosted by Institute for Physics of Intelligence (iπ), UTokyo. Large language model agents have excelled on mathematical and coding benchmarks, but can they accelerate scientific discovery in quantum physics? In this talk, I will present a multi-agent framework that couples LLM reasoning with computational backends at increasing levels of rigor and demonstrate it through two applications: (i) co-design of quantum error-correcting codes with transversal diagonal gates via multi-agent numerical search, and (ii) formal verification of tensor network theory and MIP*=RE in Lean 4. The underlying system, TeXRA, orchestrates specialized agents and builds research and verification frameworks that connect informal physical reasoning to numerical computation and machine-checked proof.
Venue: via Zoom
Event Official Language: English
-
Seminar
Genome Language Models: From DNA Sequences to Biological Foundation Models
August 13 (Thu) 15:00 - 16:00, 2026
Minrui Chen (Ph.D. Student, Kyushu University)
Recent advances in protein language models have greatly transformed protein structure prediction, functional annotation, and biomolecular design. In contrast, genome language models aim to learn directly from DNA sequences, which represent a more upstream layer of biological information encoding genes, regulatory logic, variant effects, and evolutionary signals. In this talk, I will introduce the basic motivation and recent progress of DNA and genome language models, including DNABERT, DNABERT-2, HyenaDNA, Evo, Evo 2, and AlphaGenome. I will discuss how different model architectures and tokenization strategies address the challenges of genomic sequence modeling, such as long-range dependencies, multi-scale biological structure, and genome-scale context.
Venue: #359, 3F, Main Research Building (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
Artificial Intelligence in New Physics Electroweak Phase Transition Studies
July 16 (Thu) 15:00 - 16:00, 2026
Yang Zhang (Professor, Henan Normal University, China)
The study of electroweak phase transitions in BSM involves complex numerical calculations, large parameter spaces, and the integration of multiple computational tools. In this talk, I will review recent developments in applying artificial intelligence to new physics phase transition studies. First, I will discuss how machine learning methods can accelerate electroweak phase transition studies, including efficient evaluations of phase transition dynamics, such as bounce action calculations, and the exploration of detectable parameter regions for gravitational-wave searches. Then, I will introduce the emerging role of AI agents in scientific workflows, including automated model construction, effective potential generation, and parameter scans. These developments illustrate how AI can transform traditional computational pipelines and provide new possibilities for future high-energy physics research.
Venue: via Zoom
Event Official Language: English
-
Seminar
Sampling Lattice Gauge Theories with Gauge-Equivariant Neural Networks and Diffusion Models
July 8 (Wed) 14:00 - 15:30, 2026
Andreas Ipp (Senior Scientist, Institute for Theoretical Physics, Technische Universität Wien, Austria)
Thomas Ranner (Project Assistant (FWF), Institute for Theoretical Physics, Technische Universität Wien, Austria)Simulating non-Abelian lattice gauge theories using classical Markov chain Monte Carlo (MCMC) methods is often bottlenecked by critical slowing down and topological freezing. Incorporating physical symmetries into machine learning architectures offers a powerful way towards overcoming these computational limitations. In this talk, we present an overview of how gauge-equivariant neural networks can enhance lattice simulations. In the first part, we introduce Lattice Gauge Equivariant Convolutional Neural Networks (L-CNNs), which build local gauge symmetry directly into the network structure. We discuss their ability to learn Wilson loops and highlight their versatility across different applications, such as approximating effective actions. In the second part, we focus on their application in generative modeling: gauge-equivariant diffusion models. Using a Metropolis-adjusted annealed Langevin scheme, these models are designed to generate uncorrelated field configurations. We demonstrate accurate sampling of 2D U(2) and SU(2) theories, showcase recent advancements in scaling up to 4D SU(3) pure gauge theory, and show that these models extrapolate remarkably well to larger lattices and inverse couplings beyond their training regime.
Venue: #445--447, 4F, Main Research Building (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
Machine-learned fixed-point actions and observables for SU(3) lattice gauge theory
June 24 (Wed) 10:30 - 11:30, 2026
Müller David (Postdoctoral Researcher, Institute for Theoretical Physics, TU Wien, Austria)
Lattice regularization is the established approach for studying non-perturbative phenomena in quantum chromodynamics, but accurate predictions for the continuum theory remain challenging because standard actions exhibit large lattice artifacts. The renormalization group on the lattice provides a way of suppressing these artifacts: classically perfect fixed-point (FP) actions. In this talk, I show how gauge-equivariant neural networks yield accurate parametrizations of FP actions. Using these machine-learned actions, we perform Monte Carlo simulations to measure gradient-flow scales with highly suppressed artifacts compared to unimproved actions. I will also present preliminary results for machine-learned FP observables to improve the extraction of the topological susceptibility in four-dimensional SU(3) gauge theory.
Venue: Seminar Room #359 (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
DeepQuark: A Deep-Neural-Network Approach to Multiquark Bound States
June 4 (Thu) 15:00 - 16:00, 2026
Wei-Lin Wu (Ph.D. Student, School of Physics, Peking University, China)
Recent discoveries of multiquark candidates have opened a new frontier in hadron spectroscopy and nonperturbative QCD. Understanding these multiquark states poses a challenging quantum many-body problem governed by SU(3) color interactions. Traditional approaches based on basis expansions often encounter severe bottlenecks as the system size and dynamical complexity increase. In this talk, I will present DeepQuark, a deep-neural-network-based variational Monte Carlo framework for solving multiquark bound states. I will discuss the general methodology behind neural-network quantum states, the challenges of extending existing approaches from electronic and nuclear systems to hadron physics, and the architecture of DeepQuark. By combining physics-informed symmetry constructions with the expressive power of deep neural networks, DeepQuark provides a scalable framework for studying multiquark spectroscopy and exploring confinement dynamics.
Venue: via Zoom
Event Official Language: English
-
Seminar
Generative diffusion model with inverse renormalization group flows
June 2 (Tue) 14:00 - 15:00, 2026
Kanta Masuki (Ph.D. Student, Graduate School of Science, The University of Tokyo)
Diffusion models have recently emerged as one of the most powerful frameworks for generative modeling, achieving remarkable success in a wide range of domains, including image generation, audio synthesis, and scientific data generation. However, despite their empirical success, conventional diffusion models often require many denoising steps and do not explicitly exploit the multiscale structure naturally present in various types of data. This limitation motivates us to ask whether ideas from the renormalization group (RG), which is designed to describe scale-dependent effective degrees of freedom, can provide a useful principle for constructing more efficient generative models. In this talk, I will present our recent work on renormalization-group diffusion models (RGDMs) [1], a generative framework that connects diffusion models with RG flows. By establishing a correspondence between diffusion dynamics and exact RG flow equations, we construct a diffusion model whose reverse process generates data in a coarse-to-fine manner, thereby effectively reversing an RG flow. I will first introduce the theoretical formulation of RGDMs and explain how the RG perspective leads to a coarse-to-fine generative process. I will then present numerical results in protein structure prediction and image generation, where RGDMs improve sample quality and/or sampling efficiency compared with conventional diffusion models. Finally, I will discuss possible extensions and open questions, including broader applications of RG-inspired generative modeling.
Venue: Seminar Room #359
Event Official Language: English
-
Seminar
Stochastic Schrödinger Diffusion Models for Pure-State Ensemble Generation
May 14 (Thu) 14:30 - 15:30, 2026
Jian Xu (Postdoctoral Researcher, Quantum Mathematical Science Team, Division of Applied Mathematical Science, RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS))
In quantum machine learning (QML), classical data are often encoded as quantum pure states and processed directly as quantum representations, motivating \emph{representation-level generative modeling} that samples new quantum states from an underlying pure-state ensemble rather than re-preparing them from perturbed classical inputs. However, extending \emph{score-based} diffusion models with well-defined reverse-time samplers to quantum pure-state ensembles remains challenging, due to the non-Euclidean geometry of the complex projective space $\mathbb{CP}^{d-1}$ and the intractability of transition densities. We propose \emph{Stochastic Schr\"odinger Diffusion Models} (SSDMs), an intrinsic score-based generative framework on $\mathbb{CP}^{d-1}$ endowed with the Fubini--Study (FS) metric. SSDMs formulate a forward Riemannian diffusion with a stochastic Schr\"odinger equation (SSE) realization, and derive reverse-time dynamics driven by the Riemannian score $\nabla_{\mathrm{FS}} \log p_t$. To enable training without analytic transition densities, we introduce a local-time objective based on a local Euclidean Ornstein--Uhlenbeck approximation in FS normal coordinates, yielding an analytic teacher score mapped back to the manifold. Experiments show that SSDMs faithfully capture target pure-state ensemble statistics, including observable moments, overlap-kernel MMD, and entanglement measures, and that SSDM-generated quantum representations improve downstream QML generalization via representation-level data augmentation.
Venue: #359, Seminar Room #359 (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
Building autonomous AI physicists for frontier physics research
April 30 (Thu) 15:00 - 16:00, 2026
Tingjia Miao (Ph.D. Student, School of Artificial Intelligence, Shanghai Jiao Tong University, China)
Advances in LLMs have led to agents with knowledge and operational capabilities comparable to human scientists, suggesting potential to assist, accelerate, and automate research. Physics, especially theoretical and computational physics, which requires integrating analytical reasoning, code-based computation, and profound domain expertise, is well suited for verifying the end-to-end research capabilities of AI scientists. Accordingly, we construct a general-purpose AI physicist PhysMaster, equipped with a layered academic knowledge base, adapted to the agent skill ecosystem, and adopting an adaptive exploration strategy that balances efficiency and exploration, enabling robust performance in ultra-long-horizon tasks; PhysMaster has been open-sourced. Meanwhile, we introduce PRL-Bench (Physics Research by LLMs), a benchmark with 100 tasks adapted from recent Physical Review Letters papers, covering astrophysics, condensed matter physics, high-energy physics, quantum information, and statistical physics. Evaluation across frontier models shows that failures are dominated by conceptual and formulaic errors, and that exploration and derivations remain unstable over long horizons. In addition, we develop domain-specialized AI scientists, including LQCD Master, which integrates Lattice QCD workflows and expert skills, enabling automated generation and submission of lattice computation scripts from concise physics goals.
Venue: via Zoom
Event Official Language: English
-
Seminar
DEEP-IN WG Sarter Meeting 2026
April 27 (Mon) 15:30 - 17:00, 2026
Tae-Geun Kim (Postdoc, Fudan University, China)
Yang-Yang Tan (Postdoctoral Researcher, Institute for Physics of Intelligence, Graduate School of Science, The University of Tokyo)
Masato Taki (Associate Professor, Graduate School of Artificial Intelligence and Science, Rikkyo University)15:30–16:00 NOW&NEXT of DEEP-IN WG (Lingxiao Wang) Self-Introduction of Members 16:00–16:20 AI Team and DEEP-IN (Masato Taki) 16:20–16:40 Inverse Problems in HEP (Tae-Geun Kim, FudanU) 16:40–17:00 Inverse Modeling Distributions (Yang-yang Tan, UTokyo)
Venue: #359, Seminar Room #359 (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
Quantum generative learning via diffusion
April 21 (Tue) 10:00 - 11:00, 2026
Zhang Bingzhi (PostDoc, University of Southern California, USA)
Deep generative models are key-enabling technology to computer vision, text generation, and large language models. Generative models for quantum data offer a promising route toward learning and preparing complex quantum-state ensembles. In this talk, I will introduce the quantum denoising diffusion probabilistic model (QuDDPM) [1], which adapts the diffusion-model idea to quantum systems through a forward randomization process and a trainable backward denoising dynamics. I will discuss how this framework enables stepwise learning of target quantum state ensembles and demonstrate its capabilities in various learning tasks. I will then present its extension to mixed states to eliminate the need for scrambling [2]. I will conclude with a brief discussion of recent results on scaling laws of quantum information lifetime in monitored quantum dynamics, emphasizing how mid-circuit measurements can maintain information and provide useful intuition for measurement-assisted quantum machine learning.
Venue: via Zoom
Event Official Language: English
-
Seminar
Searching For Anomalies with Foundation Models
April 16 (Thu) 14:00 - 15:30, 2026
Vinicius Massami Mikuni (Associate Professor, Kobayashi-Maskawa Institute for the Origin of Particles and the Universe (KMI), Nagoya University)
This is a joint seminar with Institute for Physics of Intelligence (iπ), UTokyo Anomaly detection relaxes the assumptions of how new physics should look and extends the reach of what we can discover. However, interpreting the data and estimating backgrounds remains a challenge. In this new work, we investigate anomalous events selected by the OmniLearned Foundation model across different model sizes, performing a full analysis using CMS Open Data. Surprisingly, models of different sizes, trained on the same data with the same loss functions, select entirely different collisions. In particular, the large OmniLearned model (500M parameters) selects events that are not well described by our background model.
Venue: Faculty of Science Bldg.1, School of Science, The University of Tokyo, Hongo Campus (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
DEEP-IN-iPI Joint Meeting
January 26 (Mon) - 30 (Fri) 2026
Xingyu Guo (Lecturer, Institute of Quantum Matter, South China Normal University, China)
Gert Aarts (Professor, Department of Physics, Swansea University, UK)
Shuzhe Shi (Assistant Professor, Physics Department, Tsinghua University, Beijing, China)
Sung Hak Lim (Senior Researcher, Center for Theoretical Physics of the Universe (CTPU-PTC), Institute for Basic Science (IBS), Republic of Korea)
Jinyang Li (Ph.D. Student, Program of Particle and Nuclear Physics, The Graduate University for Advanced Studies (SOKENDAI))
Yingying Xu (Research fellow, Department of Mathematics and Statistics, University of Helsinki, Finland)The series of DEEP-IN meetings (Jan 26–30, 2026) are joint with UTokyo Institute for Physics of Intelligence (iπ), which is a multi-day scientific program bringing together researchers to explore quantum simulations, machine learning physics, and applications in particle and nuclear physics. The tentative schedule is, UTokyo-iπ Session (Venue: #512, Faculty of Science Bldg.1, School of Science, UTokyo) Day 1: Jan 26 (Mon) 14:30–16:00 Onset of Bjorken flow in a quantum many-body simulation of the massive Schwinger model, Shuzhe Shi Day 2: Jan 27 (Tue) 14:30–16:00 Physics of Diffusion Models, Gert Aarts 16:00–17:30 Discovering Symmetry from Energy-Based Diffusion Models, Jinyang Li RIKEN-iTHEMS Session (Venue: Seminar Room #359, Main Research Building) Day 3: Jan 28 (Wed) 14:30–16:00 Understanding Galactic Dark Matter with Generative Models, Sung Hak Lim 16:00–18:00 Free Discussion ML Physics-1 Day 4: Jan 29 (Thu) 10:00–11:30 Quantum Simulations of HEP and Beyond, Xingyu Guo 14:30–16:00 Physics of Machine Learning, Gert Aarts 16:00–17:30 Storage capacity of perceptron with variable selection, Yingying Xu Day 5: Jan 30 (Fri) 11:00–14:00: Free Discussion ML Physics-2
Venue: via Zoom / Seminar Room #359 / Faculty of Science Bldg.1, School of Science, The University of Tokyo, Hongo Campus
Event Official Language: English
-
Seminar
Neural Network Quantum States for Quarkonium in Medium: Real-Time Open Quantum-System Dynamics
December 19 (Fri) 16:00 - 17:00, 2025
Tom Magorsch (Ph.D. Student, Department of Physics, Technical University of Munich, Germany)
Many phenomena in high energy physics can not be described by Euclidean-time Monte Carlo estimates alone, but require genuine real-time evolution and a treatment of non-equilibrium effects. However, such simulations are computationally challenging. One such example is the evolution of heavy quarkonium in the quark gluon plasma produced in heavy-ion collisions. In this talk, I will introduce the open quantum system treatment of in-medium quarkonium. I will then give an overview on neural network quantum states as a variational approach to the real-time simulation of open quantum systems. As a controlled benchmark system, I will study the application to the Caldeira-Leggett model and conclude with an outlook on future applications of neural network based simulation of quarkonia in medium.
Venue: via Zoom
Event Official Language: English
-
Seminar
A bi-fidelity Asymptotic-Preserving Neural Network approach for multiscale kinetic problems
December 17 (Wed) 11:00 - 12:00, 2025
Liu Liu (Assistant Professor, Department of Mathematics, The Institute of Mathematical Sciences, The Chinese University of Hong Kong, Hong Kong)
In this talk, we will introduce a bi-fidelity Asymptotic-Preserving Neural Network (BI-APNNs) framework, designed to efficiently solve forward and inverse problems for the linear Boltzmann equation. Our approach builds upon the previously studied Asymptotic-Preserving Neural Network (APNNs), which employs a micro-macro decomposition to handle the model’s multiscale nature. We specifically address a bottleneck in the original APNNs: the slow convergence of the macroscopic density in the near fluid-dynamic regime. This strategy significantly accelerates the training convergence as well as improves the accuracy of the forward problem solution, particularly in the fluid-dynamic limit. We show several numerical experiments on both linear Boltzmann and the Boltzmann-Poisson system that this new BI-APNN method produces more accurate and robust results for forward and inverse problems compared to the standard APNNs. This is a joint work with Zhenyi Zhu and Xueyu Zhu.
Venue: Seminar Room #359 (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
Generative sampling with physics-informed kernels
December 8 (Mon) 14:00 - 15:00, 2025
Renzo Kapust (Ph.D. Student, Institute for Theoretical Physics, University Heidelberg, Germany)
We construct a generative network for Monte-Carlo sampling in lattice field theories and beyond, for which the learning of layerwise propagation is done and optimised independently on each layer. The architecture uses physics-informed renormalisation group flows that provide access to the layerwise propagation step from one layer to the next in terms of a diffusion equation for the respective renormalisation group kernel through a given layer. Thus, it transforms the generative task into that of solving once the set of independent and linear differential equations for the kernels of the transformation. As these equations are analytically known, the kernels can be refined iteratively. This allows us to structurally tackle out-of-domain problems generally encountered in generative models and opens the path to further optimisation. We illustrate the practical feasibility of the architecture within simulations in scalar field theories.
Venue: Seminar Room #359 (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
Hamiltonian Learning and Dynamics Prediction via Machine Learning
November 26 (Wed) 15:00 - 16:00, 2025
Li Keren (Assistant Professor, College of Physics and Optoelectronic Engineering, Shenzhen University, China)
Accurate prediction of quantum Hamiltonian dynamics and identification of Hamiltonian parameters are crucial for advancements in quantum simulations, error correction, and control protocols. This talk introduces a machine learning model with dual capabilities: it can deduce time-dependent Hamiltonian parameters from observed changes in local observables within quantum many-body systems, and it can predict the evolution of these observables based on Hamiltonian parameters. The model’s validity was confirmed through theoretical simulations across various scenarios and further validated by two experiments. Initially, the model was applied to a Nuclear Magnetic Resonance quantum computer, where it accurately predicted the dynamics of local observables. The model was then tested on a superconducting quantum computer with initially unknown Hamiltonian parameters, successfully inferring them. We believe that machine learning techniques hold great promise for enhancing a wide range of quantum computing tasks, including parameter estimation, noise characterization, feedback control, and quantum control optimization.
Venue: via Zoom
Event Official Language: English
-
Seminar
On the Role of Hidden States of Modern Hopfield Network in Transformer
November 10 (Mon) 14:00 - 15:00, 2025
Masato Taki (Associate Professor, Graduate School of Artificial Intelligence and Science, Rikkyo University)
Large language models such as ChatGPT are based on deep learning architectures known as Transformers. Owing to their remarkable performance and broad applicability, Transformers have become indispensable in modern AI development. However, it still remains an open question why Transformers perform so well and what the essential meaning of their unique structure is. One possible clue lies in the mathematical correspondence between Hopfield Networks and Transformers. In this talk, I will first introduce the major developments over the past decade that have significantly increased the storage capacity of Hopfield Networks. I will then review the theoretical correspondence between Hopfield Networks and Transformers. Building on this background, I will present our recent findings: by extending this correspondence to include the hidden-state dynamics of Hopfield Networks, we discovered a new class of Transformers that can recursively propagate attention-score information across layers. Furthermore, we found, both theoretically and experimentally, that this new Transformer architecture resolves the “rank collapse” problem often observed in conventional multi-layer attention. As a result, when applied to language generation and image recognition tasks, it achieves performance surpassing that of existing Transformer-based models.
Venue: Seminar Room #359, Seminar Room #359 (Main Venue) / via Zoom
Event Official Language: English
-
Seminar
Quantum multi-body problems using unsupervised machine learning
November 5 (Wed) 15:00 - 16:00, 2025
Tomoya Naito (Project Assistant Professor, Department of Nuclear Engineering and Management, Graduate School of Engineering, The University of Tokyo)
I will introduce the recent development of a method to calculate the (anti)symmetrized wave functions and energies of the ground and low-lying excited states using the unsupervised machine learning technique. I will also introduce the recent attempts to consider the spin-isospin degrees of freedom and extend them to the Dirac equation.
Venue: Seminar Room #359 (Main Venue) / via Zoom
Event Official Language: English
42 events