Academia-Industry Matching Event (AIME26)

→ Europe/Budapest
HUN-REN Headquarters

HUN-REN Headquarters

1054 Budapest Alkotmány utca 29.
Description

HUN-REN Wigner RCP, together with the HEPTECH Network, is organizing the next

Academia-Industry Matching Event (AIME26)

in the topics

Artificial Intelligence, HPC, and Quantum Computing

The aim of this event is to bring together Academic researchers and Industry experts to share ideas, potential applications, and foster collaborations in the field of theoretical and practical aspects of Artificial Intelligence, High-Performance Computing (HPC), and Quantum Computing.

Earlier events: 2025, 2024, 2023, 2022, 2021, 2019, 2018

Sponsors

Patron

  • István Csabai (ELTE, Professor)
  • Péter József Lévai (HUN-REN Wigner RCP, Director General)

 

Confirmed keynote speakers

  • Örs Legeza, HUN-REN Wigner Research Centre for Physics - Bridging Classical and Quantum Computing via Advanced Tensor Network Simulations on High-Performance Computing Infrastructures
  • Sanja Damjanovic, CERN, GSI Helmholtzzentrum für Schwerionenforschung  - From Heavy-Ion Physics to Societal Impact: The SEEIIST Initiative
  • Gianluca Inguglia, Marietta Blau Institute for Particle Physics of the ÖAW - From Gravitational Waves to Artificial Intelligence

 

We warmly welcome abstract submissions for talks from all of our participants and look forward to hearing about your work!

Please note that abstracts can be submitted only through an Indico account. If you work at an institution that is a member of the Shibboleth federation, you can sign in quickly using your institutional credentials:

  1. Click “Login with EduID”.
  2. Start typing the name of your institution (e.g., Wigner) and select it from the list.
  3. Authenticate with your institutional account and submit your abstract.

 

If your institution is not part of Shibboleth, you are very welcome to create an account in our Indico system. Alternatively, you may simply email us the title of your talk, the abstract, the author list, and the name of the speaker, and we will gladly take care of the submission for you.

Each abstract should include a title of no more than 100 characters, the names and affiliations of the presenter and all co-authors, and a main text of up to 4,000 characters. As abstracts will be reformatted for display on the website, please send any images separately from the text.

We are excited to receive your contributions!

 

Thanks to the generous support of HEPTECH, registration is free for all participants! We also warmly welcome sponsors who’d like to showcase their work. Sponsorship packages start at just 200 EUR and can include exhibition space, a dedicated desk, and other opportunities to connect with attendees.

Talk submission deadline: November 8, 2026

The Organizing Committee:
Gergely Gábor Barnaföldi
Gábor Bíró
Balázs Kacskovics
Orsolya Kálmán
Péter József Lévai
András Telcs

 

Participants
  • Thursday 12 November
    • Opening: Welcome & Opening
      Conveners: Dr Gergely Barnafoldi (HUN-REN Wigner Reseach Centre for Physics), Peter Levai (WIGNER RCP)
      • 1
        Welocme and Opening
    • Artificial Intelligence
      Convener: Peter Levai (WIGNER RCP)
      • 2
        From Gravitational Waves to Artificial Intelligence

        Since the first experimental detection of gravitational waves in 2015, observations in this field have profoundly transformed our understanding of the Universe.
        In this presentation, I will offer an overview of gravitational wave science, emphasising how this field naturally operates at the interface of multiple disciplines, ranging from signal processing to astrophysics to artificial intelligence. I will review the current status and prospects of detectors such as LIGO–Virgo–KAGRA and the Einstein Telescope, and discuss how their observations will enable deeper insights into black hole populations, neutron-star mergers, and the early history of the Universe.
        Finally, I will address how the increasingly central role of artificial intelligence, machine learning, and agentic workflows in gravitational-wave science is changing the way we do research, from managing massive data streams to improving signal detection and extracting physical parameters.

        Speaker: Gianluca Inguglia (MBI Vienna - Austrian Academy of Sciences)
      • 3
        Generative AI Meets Neuroscience: Learning, Uncertainty, and Adaptation in Visual Perception

        Despite remarkable advances in artificial intelligence, modern machine learning systems still face important challenges, including reliable uncertainty estimation, robustness to distribution shifts, and efficient adaptation to new tasks. Biological intelligence offers a valuable source of inspiration for addressing these limitations.

        In this talk, I will explore the connections between generative AI and neuroscience, focusing on how the brain combines sensory evidence with prior knowledge to interpret the visual world. I will present our research on brain-inspired generative models, particularly Variational Autoencoders (VAEs), that improve uncertainty estimation, capture hierarchical information processing, and enable efficient adaptation to changing contexts.

        Beyond drawing inspiration from biology, these models also provide computational tools for understanding neural activity. In collaboration with experimental neuroscientists, we have demonstrated how context-dependent responses in the mouse visual cortex can be explained through probabilistic inference with adaptable priors.

        Our work highlights the potential of a two-way exchange between AI and neuroscience: developing more robust and adaptive machine learning systems while gaining a deeper understanding of biological intelligence.

        Speaker: Balázs Meszéna (Wigner RCP)
      • 4
        How to solve IQ tests with a miniature AI.

        The lecture is a progress report. The foundations and work is presented of the mAI. The mAI is based on a knowledge space (KS), a local, deterministic solver and the controller of the KS extension. The KS initially equipped with a small set of concepts, relations and cognitive primitives. The solver is limited to the cognitive primitives (later schemas, procedures, operations) and concepts stored in the KS but can combine them to find solutions. New concepts, procedures used in a successful solution proposed to be inserted to the KS for reuse. The KS evolves as tasks are provided to the mAI.

        Speaker: Andras Telcs
    • SciComp Presentation
      Convener: Andras Telcs
    • 10:45
      Coffee Break
    • Artificial Intelligence
      Convener: Andras Telcs
      • 5
        Software Sustainability Index

        Software Sustainability Index

        Speaker: Róbert Ceglédi (Neuron Software Ltd.)
      • 6
        Auditable LLM Pipelines for Text Measurement: Conceptual Frame Profiling

        Policy documents do not just report the economy, they describe it through metaphor. The minutes of the U.S. Federal Reserve’s rate-setting committee (FOMC) say that inflation is “climbing” or that commercial paper issuance “almost dried up”. In the first case, an abstract phenomenon, inflation (the target domain), is understood in terms of a concrete experience, vertical movement (the source domain). Together they form the conceptual metaphor INFLATION IS VERTICAL MOVEMENT. In the second, the financial market (target) is understood as a liquid (source): FINANCIAL MARKET IS A LIQUID. The set of metaphors a text relies on is a partial window on its conceptual frame, the system of concepts through which its authors understand the economy. Standard sentiment tools measure how negative a text is, but not which frames it uses. We present Conceptual Frame Profiling (CFP), an LLM-based pipeline that identifies metaphorical expressions and aggregates them into quantitative, document-level frame profiles, and we evaluate it as a measurement instrument.

        CFP has three steps. (i) Gemini 2.5 Pro finds metaphorical expressions and classifies them under a fixed, schema-constrained codebook: 9 source domains (e.g., liquid, machine, vertical movement), 5 target domains (e.g., inflation, financial market), and 3 graded states (1 = under control, 2 = tension, 3 = loss of control). (ii) Because LLM output is stochastic, we repeat the extraction three times and keep only expressions confirmed across runs. (iii) The retained expressions are aggregated per document into a frame profile (the shares of the source domains) and a strain index, the mean state of the document’s expressions (range 1–3). It measures how strained the description is, not how strained the economy is.

        We applied CFP to 257 FOMC minutes (1992–2026, ~1.76 million words). 76.3% of the identified expressions are reproduced by an independent run. Three annotators judged 84.5–92.8% of the expressions in fifty sampled passages correct, and agreed with each other on source domain at κ = 0.82–0.91 (chance-corrected agreement). A second, open-weights model (Gemma 4) reproduces the source domain classification (κ = 0.82) and the document-level profiles (r = 0.89), so the method does not depend on one proprietary model.

        Compared with two standard sentiment measures (Loughran–McDonald dictionary and FinBERT), the strain index partly overlaps with sentiment (r = −0.66 and −0.52). The frame composition, however, is nearly independent of it: sentiment explains at most 6.4% of the variance of any source domain share, and crisis phases still differ from normal periods after controlling for sentiment. In an exploratory analysis, the profiles are concentrated in four source domains, stable over three decades, and shift at the boundaries of historically labelled financial-cycle phases (boom, panic, etc.) together with four financial stress indices.

        The codebook, prompts, run outputs, and code are released. The approach is domain-agnostic and could be applied to other large collections of expert text, such as regulatory communication or financial disclosures.

        Speaker: Orsolya Putz
      • 7
        Causal Methods in High-Dimensional Data Analysis

        High-dimensional time series and cross-sectional data are increasingly common in finance, economics, medicine, physics, and many other fields. Building predictive models for such systems is challenging because of the large number of potential input variables. Including too many variables may lead to unnecessarily complex models, reduced predictive performance, and difficulties in interpretation.

        Identifying the variables that are genuinely relevant to the underlying dynamics is therefore a crucial step in model construction. Causal methods provide a powerful framework for this task, allowing us to distinguish meaningful dependencies from spurious correlations and to identify the most relevant variables.

        In this talk, I will discuss how different types of causal relationships can be uncovered directly from observational data, using methods that do not require an explicit model of the underlying system.

        Speaker: Antal Jakovac (Wigner RCP, Department of Computational Sciences)
    • 12:45
      Group photo
    • 12:50
      Lunch
    • High-performance Computing
      • 8
        From Heavy-Ion Physics to Societal Impact: The SEEIIST Initiative

        Heavy-ion therapy is one of the most successful applications of nuclear and particle physics, translating decades of research in heavy-ion accelerators, beam physics, and ion–matter interactions into clinical practice.
        Pioneering work at the GSI Helmholtz Centre for Heavy Ion Research, where the first patients in Europe were treated with carbon ions, laid the foundations for the establishment of the Heidelberg Ion-Beam Therapy Center. Contributions from CERN to accelerator technologies also supported the creation of the National Center for Oncological Hadrontherapy in Pavia (CNAO). Carbon-ion therapy exploits the unique physical and radiobiological properties of heavy ions to treat challenging cancers with exceptional precision.
        Building on these achievements, this contribution will present the SEEIIST initiative (South East European International Institute for Sustainable Technologies), a next-generation carbon-ion therapy and biomedical research facility based on a novel medical accelerator. Particular attention will be given to recent advances in accelerator science, including superconducting magnet technologies, and emerging approaches to particle therapy, such as ultra-high dose-rate (FLASH) irradiation.
        Beyond its scientific and medical objectives, SEEIIST aims to establish a shared research infrastructure for South-East Europe, combining advanced cancer treatment and biomedical research with technological innovation, regional capacity building and international scientific cooperation.

        Speaker: Sanja Damjanovic
      • 9
        CT Imaging with Hadrons for Cancer Radiotherapy

        Hadron therapy is an emerging form of cancer treatment that shows to be more effective than the most widespread type of radiotherapy treatment with X-rays, due to the nature of the energy deposition of hadrons. Before the treatment, there is a need for the imaging of the patient, which is usually done by using X-ray CT. However, with the usage of the conventional X-ray CT, a conversation of the Hounsfield units (describing the photon's absorption) to Relative Stopping Power values (describing how fast the hadrons stop in different materials) is necessary and introduces ~7% error. A possible solution for eliminating the uncertainty could be imaging with the same particle that is used for the treatment later.
        This work focuses on the comparative analysis of proton, helium-ion, and carbon-ion computed tomography (CT) as direct imaging modalities for hadron therapy treatment planning, focusing on Relative Stopping Power (RSP) reconstruction accuracy and patient radiation dose. For this, high-fidelity Monte Carlo simulations were conducted using the GATE/Geant4 platform to model a standard CTP404 phantom. RSP maps were reconstructed using two iterative algorithms: the Simultaneous Algebraic Reconstruction Technique (SART) and the Richardson-Lucy (RL) deconvolution. Imaging performance was evaluated by comparing reconstructed RSP values against ground truth data for various tissue-equivalent inserts, while integral doses were estimated for a human head geometry using a Shepp-Logan-like phantom.
        All investigated particle modalities demonstrated a significant dose reduction compared to conventional X-ray CT protocols (which are approximately 40 mGy). The estimated imaging doses were $0.5$ mGy for protons, $1.3$ mGy for helium ions, and $7.5$ mGy for carbon ions.
        In terms of accuracy, carbon-ion imaging achieved promising accuracy for soft-tissue materials (mean absolute error $<0.5\%$). Helium ions offered a balanced performance with sub-$1\%$ errors for most materials and a dose burden significantly lower than carbon ions. Protons exhibited the widest range of RSP deviations.
        Direct particle imaging eliminates the systematic uncertainties inherent in photon-to-hadron conversion, which typically introduce RSP deviations of 1.6\%–3.5\% in standard single-energy X-ray CT. While carbon ions provide better RSP reconstruction precision essential for complex treatment plans in heterogeneous anatomy, helium and proton imaging offer exceptional dose sparing, making them particularly advantageous for pediatric patients and frequent adaptive replanning scenarios.

        Speaker: Zsófia Jólesz
    • 15:35
      Coffee Break
    • High-performance Computing
      • 10
        Scientific Reliability of AI Tools in 3D Bioinformatics and Protein Engineering

        Our objective is to develop an flexible in silico pipeline for designing small protein binders targeting an immune checkpoint receptor, aiming to overcome immune suppression and enhance cancer cell killing. Such protein design pipelines are complex, integrating multiple AI-based methods, including diffusion models, deep neural networks, and protein structure prediction tools.

        We discuss the evaluation of existing protein design pipelines and the development of a customized workflow, supported by both human experts and the AI coding agent Codex. Particular attention is given to scientific, logical, and numerical errors identified in individual components of these pipelines. Some errors produce clearly incorrect results that compromise downstream predictions, while others introduce numerical inaccuracies without necessarily affecting the biological interpretation of the results. Distinguishing between these categories is essential for assessing the reliability of computational predictions.

        We argue that scientific software, particularly AI-based tools, should undergo standardized quality assessments that extend beyond conventional software testing to include scientific validity, algorithmic correctness, and numerical reliability. AI coding agents are increasingly capable of identifying scientific and logical flaws in complex computational workflows and could play an important role in establishing reproducible, systematic quality control in computational biology.

        Speaker: Tamas Hegedus (Semmelweis University)
      • 11
        Towards Dedicated AI Infrastructure at Wigner RCP: Motivations and Challenges

        The development and operation of private Artificial Intelligence (AI) capacity represents a strategic opportunity for research institutions and R&D-intensive companies to strengthen scientific excellence, technological independence, and innovation competitiveness.

        Dedicated AI infrastructure enables secure processing of sensitive research data, development of domain-specific models, and acceleration of computational experiments, simulations, and scientific discoveries. Beyond computational resources, private AI capacity supports the creation of intellectual property, advanced research methodologies, and industrial partnerships.

        In this context, I would like to discuss the necessity of investing in new AI hardware capacity and the key challenges associated with moving in this direction, including infrastructure costs, technological evolution, resource utilization, energy efficiency, and long-term sustainability. The discussion will explore strategic motivations, research opportunities, and potential approaches to developing sustainable and competitive private AI ecosystems.

        Speaker: Ádám Pintér
      • 12
        The "Navier-Stokes Equation" Millennium Prize Problem: Physics behind Mathematics

        In this lecture, we will explore the Clay Institute's Millennium Prize Problem concerning the Navier–Stokes equations. First, we'll review its origins, starting with the sixth of Hilbert's famous 23 problems, moving through the Lax–Friedrichs theorem to the mathematical role of thermodynamic concepts. Then I'll show exactly what needs to be solved from a mathematical perspective, what OpenAI's team of AI agents has solved, and what they haven't. It will become clear where the Navier–Stokes equations originate and why we expect that the solution to a dissipative equation, under smooth initial conditions, should not explode on its own or give rise to shock waves and the like. However, from a physical perspective, the mathematical problem is strangely degenerate, and its difficulty is truly revealed by the energy balance that necessarily accompanies the momentum balance. Finally, in this context, we might reflect on what actually happened when mathematicians refer to the "Kasparov vs. Deep Blue" moment in mathematics, and wonder when such a moment might arrive in physics. Looking back at the events and the debate over priority, it is certain that we are nowhere near the point where our common sense would become useless, and we can ponder how we should use it differently than we have so far.

        Speaker: Peter Van (Wigner FK, RMI, Elméleti Fizika Osztály)
  • Friday 13 November
    • Quantum Computing and Technology
      • 13
        Classically trainable quantum generative modeling

        Sampling-based quantum advantage schemes currently serve as the leading experimental demonstrations of computational tasks deemed intractable for classical devices. However, their practical significance remains a subject of intense debate as these tasks are synthetically designed to maximize classical hardness rather than to solve meaningful real-world problems. In this contribution, based, we propose a generative learning application of such schemes: Fermion Sampling and (Gaussian) Boson Sampling. We show that, despite the classical hardness of sampling from their output distributions, the generative models built on these schemes can nevertheless be trained efficiently by classical means.

        Speaker: Bence Bakó (Wigner RCP)
      • 14
        Bridging Classical and Quantum Computing via Advanced Tensor Network Simulations on High-Performance Computing Infrastructures

        light of emergent evolution in quantum technology, key industrial players allocate significant parts of their budget and resources to identify a real-world problem where quantum advantage, i.e. exponential increase in computational capacity, is expected to appear. The tense competition between quantum computing and simulations on classical hardware over the past decades has been further accelerated by the sever need to develop cost-effective technologies to manipulate things on the atomic and nano scale and control processes that could lead to new products to improve everybody’s life. Currently, Graphics Processing Units (GPUs) accelerated hardware provide unprecedented computational power leading to paradigmatic changes in various business cases. In this contribution, we will overview a class of quantum-inspired algorithms which increase the efficiency of classical computing algorithms without requiring any quantum hardware, and which at the same time are also capable to fully utilize state-of-the-art classical hardware available on high performance computing infrastructures. We will also show-case recent results via classical and quantum computing achieved in collaboration with NVIDIA, IBM and Google startup SandboxAQ addressing important questions like scalability, energy demand and sustainability. Key aspects of transferring academic research to industrial world as part of productization will also be addressed. Finally, we argue that for such Blue Ocean Strategy the required technology is around us, thus raising related R&D to a professional stage would revolutionize future technologies in image recognition, data process, and AI among many others, and open entire new development directions in pharma industry, materials science and beyond.

        Speaker: Örs Legeza (Wigner FK)
      • 15
        Quantum Machines – Advancing Quantum Computing Through Innovation and Engineering

        Quantum Machines is a global leader in quantum control technology, developing advanced hardware and software solutions that enable the operation and scaling of quantum computers. Its flagship OPX1000 platform combines precision analog electronics, high-speed digital processing, FPGA technology, and sophisticated real-time control software to support a wide range of quantum computing architectures.

        The presentation will provide a brief introduction to Quantum Machines, its technologies, products, and global activities. It will also highlight the company’s long-standing engineering collaboration in Hungary and the acquisition of PCB Engineering Kft. in June 2026.

        The acquisition establishes Budapest as an important hardware research and development center within Quantum Machines, strengthening its capabilities in advanced electronics and supporting the next generation of quantum computing systems.

        Speaker: Mr János Lazányi (Quantum Machines)
    • 10:45
      Coffee Break
    • Quantum Computing and Technology
      • 16
        Faulhorn Labs: a deep tech startup studio

        In this talk, I explain how Faulhorn operates, what our goals are, and how we connect with industry, academia, and research institutes. I will highlight our past projects and spin-off companies, and detail our current work on the Csip, ZKP, MC2P and Qutility projects. Finally, I discuss how AI impacts us, how we are adopting it, and how we aim to create practical value in this new era.

        Speaker: Győző Egri (Faulhorn Zrt)
      • 17
        The Qubits of Tomorrow

        Point defects are localized irregularities in crystals that break the lattice symmetry of the hosting material. They can potentially be used in quantum technology as a room temperature qubit. Nitrogen-vacancy center (NV center) is the most studied point defect, where in the hosting diamond crystal two adjacent carbon atoms are switched to a nitrogen atom and a vacancy (empty spot).
        The research of NV center is an immense success, resulting in demonstration of its capabilities across sensing, communication and computing. In sensing, NV center based technology reached commercialization.
        While NV center is a major success due to its favorable properties, finding a similarly favorable point defect among the tens of thousands of other point defects is very difficult: The mixed theoretical and experimental approach used for NV center cannot be replicated en masse.
        In the Qutility project, we develop software to provide a much more streamlined workflow for new point defect based qubit discovery.
        We develop Quandidate, an ab-initio, wavefunction theory based software package, which has the unique capability to predict the relevant properties of a point defect or molecule, which decide if it can be engineered into a useful qubit. The prediction is based on predicting relaxation rates of the excited states of point defects and molecules.
        Our Simphony open-source software models the spin dynamics of any point defect, providing a virtual laboratory for such physical systems.
        The QUASAR package is a quantum compiler, capable of compiling logical circuits to the native gateset of central spin systems.
        We also develop Qutility Frontier, a scalable and hardware-agnostic quantum benchmarking protocol.

        Speaker: Gyorgy Feher (Faulhorn Zrt)
      • 18
        QCanvas and QCR: Open Platforms for Teaching, Sharing, and Reproducing Quantum Circuits

        As quantum computing research and teaching continue to grow, so does the material behind each result and each lesson. Reproducing a published implementation depends on having its code, the tools it was built with, and the paper that explains it together, and a single lesson often draws on several tools, from PDFs for the theory and a circuit designer such as IBM Quantum Composer to a programming environment for each SDK and slides for presenting. We present two connected, open-access platforms that bring this material together. QCanvas supports teaching and learning, and the Quantum Circuit Repository (QCR) supports discovering, reproducing, and sharing quantum circuits. Both are developed under QCR MTÜ, a non-profit association.

        QCanvas (qcanvas.org) is a browser-based interactive canvas for quantum education and collaboration. Users arrange plugin elements such as text, mathematical expressions, circuits, simulations, quizzes, hint boxes, and media on a single canvas, producing self-contained interactive lessons that an educator can share with a class and a student can work through at their own pace. Its use cases extend from the mathematics behind quantum computing, such as complex numbers, linear algebra, and probability, to designing circuits, simulating them, and running them on real hardware. It is paradigm-agnostic, supporting qubit and photonic circuits within a single interface, and circuits are simulated using Qiskit and Piquasso with results shown on the canvas as Bloch spheres, Q-spheres, and probability distributions. Through a connected qBraid account, circuits run on hosted simulators and real quantum computers. Circuits export to quantum assembly languages and intermediate representations such as OpenQASM, Quil, and QIR. They also export to popular SDKs such as Qiskit, PennyLane, CUDA-Q, Cirq, pytket, Piquasso, and PsiQuantum Workbench, connecting the canvas to the existing quantum software ecosystem. For publications, circuits and figures export to LaTeX, PNG, and SVG. Its plugin architecture lets new tools and SDKs be added as it grows. QCanvas needs no installation and already has more than 200 users at high schools, universities, and other institutions.

        QCR (qcrepository.org) is an open-access, community-based repository for quantum computing. Its over 300 entries are organized into four connected categories of ready-to-run code, tutorials, tools, and papers. A reader can start from a paper and see its different implementations, the tutorials that explain them, and the tools they use, with the information needed to reproduce the results. Entries are openly searchable by topic, framework, and category. Every entry receives a persistent QCR identifier that can be cited in publications and preserves earlier versions as the entry evolves, and entries that come from a publication also record the original DOI. In this way QCR pairs the permanent, citable references familiar from arXiv with the open, community sharing of platforms like Hugging Face.

        The two platforms are designed to complement each other, and we are integrating them so that QCanvas serves as a playground for QCR, where an entry can be opened, simulated, and explored interactively, and tutorials and algorithms designed in QCanvas can be published to QCR with a citable identifier. In this talk we present both platforms and how they are used in teaching and research. By bringing lessons, circuits, and the code and papers around them together in open and citable places, QCanvas and QCR aim to make quantum computing easier to learn, find, reproduce, and build on. We welcome academic and industry partners interested in using, extending, or integrating with both platforms.

        Speaker: Szabolcs Joczik (QCR MTÜ)
      • 19
        A story of a startup: from zero to one
        Speakers: Benedek Botár (Quantum Nexus Ltd.), Bálint Farádi (Quantum Nexus Ltd.), Kornél Kemecsei (Quantum Nexus Ltd.)
    • 13:20
      Lunch
    • Quantum Computing and Technology
      • 20
        Gibbs Samplers: Nature-inspired quantum algorithms for simulating thermalisation (XPRIZE Quantum Applications finalist team)

        XPRIZE Quantum Applications, sponsored by Google Quantum AI and GESDA, is a global competition seeking new impactful applications of future fault-tolerant quantum computers; our team is one of the 12 finalists. The classical Metropolis algorithm and Glauber dynamics originated in statistical physics, yet became foundational Markov chain Monte Carlo methods with applications far beyond their original scope. We develop quantum generalisations of these dynamics, yielding efficient algorithms for preparing Gibbs (thermal) states via simulating open-system thermalisation. As a flagship application, we target the Fermi-Hubbard model in the parameter regime relevant to high-temperature superconductivity, and demonstrate the viability of this approach.

        Speaker: András Gilyén
      • 21
        From Supercomputer to Quantum Sensor: Simulating and Measuring Spin Qubits in Tandem

        Optically addressable electron spins in solids – point defects in diamond, silicon carbide and two-dimensional materials, and molecular spins – are leading platforms for room-temperature quantum sensing. Their development benefits from a close loop between theory and experiment: first-principles simulations on high-performance computers predict structures and spin and optical fingerprints, measurements test the predictions, and the data in turn refine the models. I will illustrate this approach with two recent results. First, together with colleagues at the University of Stuttgart, we showed that pentacene molecules anchored on hexagonal boron nitride act as optically readable spin sensors with zero depth. Density functional theory calculations explain how the molecules bind upright to surface defects and why their zero-field splitting increases. Deuteration and dynamical decoupling extend the spin coherence time beyond 300 µs, and the sensors detect proton spins and a 2D magnet at room temperature. Second, our laboratory introduced pulse-free quantum relaxometry under continuous-wave excitation. It measures the spin relaxation time T₁ from 60 µs to 200 ms with a single method, and it speeds up nanodiamond-based sensing of ions in water. Finally, I will outline opportunities for academia–industry collaboration in quantum sensing and in HPC- and machine-learning-driven materials design.

        Speaker: Adam Gali (No institute)
      • 22
        Quantum Optical Model of Tin-Vacancy Electronuclear Qubits in Diamond

        The strong hyperfine splitting of electronuclear tin-vacancy (SnV) levels in diamond has been shown to allow for optically addressable qubits with long coherence times. In this work we present a model of non-zero nuclear spin SnVs in diamond coupled to a quantised optical mode; the model corresponds to SnVs in photonic crystal cavities or optical cavities made of mirrors, realised in earlier experiments. We use a consistent perturbative treatment to simplify the model to describe all-optical single- and two-qubit manipulations, where the qubit manifold is defined by SnV electronuclear levels. With experimental or ab initio parameters as inputs, we are able to provide time-scale estimates for carrying out such operations. Our work has implications for quantum technological applications of SnV centres as long-lived optical qubits.

        Speaker: József Mák (Wigner Research Centre for Physics)
      • 23
        Learning to stop seizures in silico: a GPU reinforcement learning environment from neuron models

        Personalized non-invasive neurostimulation is a candidate tool for closed-loop seizure control. We ask whether a reinforcement learning (RL) agent can learn when and how strongly to stimulate to stop seizure-like activity in a model network. We wrote a network of 100 coupled Morris–Lecar neurons, extended with calcium dynamics, as a TorchRL environment. Each agent step in the environment advances the differential equations with torchdiffeq. The agent sees only a time-delay embedding of population-level measures, as an electrode would. The reward weighs seizure-free time against stimulation strength and charge balance. A PPO agent learned state-dependent multi-pulse stimulation that ended collective oscillations in a growing share of the 200 parallel network instances. We also discuss the pitfalls: reward weights and batch sizes needed careful tuning, and the GPU speed-up came almost entirely from running many copies of the network side by side. A single network simulates faster on CPU, because the time steps run one after another and each is small.

        Speaker: Marcell Stippinger (HUN-REN Wigner FK RMI Komputációs Tudományok Osztálya)
      • 24
        Who else reads your code? Mitigating data leaks when coding with AI agents

        AI coding agents are becoming part of everyday scientific work. Yet every file, prompt and test output an agent reads may leave the institute. This puts intellectual property at risk, both for new research results that we may want to patent and in joint academic–industry projects. Moreover, contracts or data protection law may forbid passing sensitive data to third parties altogether. As AI ambassador of HUN-REN Wigner RCP, I present practices that keep the benefits of AI while limiting this exposure. We run agents in isolated environments that work on both Linux and Windows. We explore how to set up devcontainers, which separate the agent from the rest of the machine and keep builds reproducible, from C++ to GPU-enabled PyTorch. We review how to work on projects that need secrets, such as access tokens and API keys. We show how to keep these secrets out of agent transcripts. We set up a VS Code environment that uses the GenAI4Science models of HUN-REN Science Cloud, so model requests stay on HUN-REN's own servers. We also point out less obvious leaks and show how the choice of agent, model and account decides whose data terms apply. The presentation includes a demo that shows the setup step by step.

        Speaker: Marcell Stippinger (HUN-REN Wigner FK RMI Komputációs Tudományok Osztálya)
    • Closing Remarks