Upcoming events

18 March 2025 16:00 till 17:00

[AN] Yuezhao Li: On robustness of topological phases of matter

Topological insulators are materials that are insulating in the bulk but allow boundary currents to flow without dissipation. These materials can be mathematically described by an invertible or projection in a C*-algebra that describes physical observables, thus classified by the K-theory of this C*-algebra. These K-theory classes are, however, not always robust under arbitrary disorder, leading to a distinction between "strong" and "weak" topological phases.

In this talk, I will explore the underlying mechanisms of the robustness of topological phases, offering insights from several perspectives, ranging from the simplest periodic models, to the more complicated groupoid and coarse-geometric techniques. Based on work of myself and a joint project together with Guo Chuan Thiang.

No preliminary knowledge is assumed for the talk. I will focus on the physical ideas. The use of K-theory and C*-algebras will only be on an intuitive level. Physicists are particularly welcome to join.

21 March 2025 12:30 till 13:15

[NA] Antoine Lechevallier : Enhancing History Matching with Machine Learning: A Hybrid Newton Approach

"History matching involves adjusting model parameters to improve the fit between observed and simulated data, ensuring better forecasting and reservoir management. Traditional ensemble-based methods, while effective, are computationally intensive, particularly for large datasets and complex reservoir models. This study introduces an innovative methodology that builds on the ""Hybrid Newton"" method by leveraging machine learning (ML) for non-linear preconditioning to accelerate history matching, with a specific focus on incremental learning for near-well prediction.

Part 1: Methodology Introduction on the Drogon Field (Rehearsal for SPE RSC25)

The iterative nature of history matching requires frequent updates to parameters such as porosity and permeability to account for uncertainties. Ensemble-based techniques refine multiple model realizations but at a high computational cost, largely due to repeated simulations. Machine learning presents a promising solution, with deep learning capturing non-linear and time-dependent dynamics. Surrogate models help reduce computational demands while maintaining accuracy, yet training these models for ensemble-based history matching remains challenging due to large datasets and generalization issues.

To address this, we enhance standard solvers with ML-based optimizations. Deep neural networks can predict reservoir outputs efficiently and improve solver performance by refining initial guesses. The Hybrid Newton method reduces the number of iterations required for convergence by leveraging ML-assisted initial guesses. Specifically, the local hybrid Newton strategy focuses on near-well regions, where well events introduce non-linear stiffness. Well dynamics often cause significant changes in reservoir behavior, necessitating drastic time-step reductions for iterative non-linear solvers to achieve convergence. Well behavior tends to exhibit similarity across spatial and temporal dimensions, presenting an opportunity for incremental learning.

We capitalize on this by training deep learning models tailored to each well’s near-well region after each history matching ensemble. These models predict solutions in the near-well region and serve as initial guesses for Newton’s method, ensuring faster convergence. Our methodology integrates ML models as local non-linear preconditioners for subsequent history matching ensembles, leveraging nested parameter distributions from prior simulations. This approach is cost-effective, as it does not require additional simulations and remains computationally feasible due to its localized application. Implemented using the open-source Python Ensemble Toolbox for ensemble modeling and OPM Flow for simulation, our methodology is validated on the Drogon field dataset. Results demonstrate a significant reduction in non-linear solver iterations required for convergence. Moreover, training deep learning models for each well can be carried out in parallel on CPUs, further enhancing efficiency.

Part 2: Assessment of SWIM on a Simplified Test Case

While the methodology has shown promising results on the Drogon field dataset, further assessment is necessary to evaluate computational efficiency and generalization capabilities. The Sample Where It Matters (SWIM) strategy has been proposed to reduce training time by optimizing data selection during supervised learning, particularly when integrating ML into solvers. High offline data generation costs for training neural networks have historically limited computational benefits, and SWIM aims to address this issue.

To validate SWIM, we employ a simplified test case with basic reservoir geometries and black-oil equations within the OPM Flow simulator. The strategy reduces computational demand by selectively training ML models where discrepancies are most pronounced, ensuring targeted learning without excessive redundant computations. Fully connected neural networks enhance Newton’s method, with SWIM ensuring efficient training on parallel CPUs. The combination of ensemble-based history matching with nested parameter distributions and SWIM optimization results in significant improvements in efficiency and solver robustness.

Conclusion

This research highlights the potential of machine learning to augment traditional history matching processes. By integrating ML-enhanced solvers and incremental learning strategies, we achieve a scalable and efficient framework that accelerates history matching while maintaining robustness. The proposed methodology can be seamlessly integrated into existing workflows, complementing conventional solvers while preserving their theoretical guarantees. Ultimately, our work paves the way for broader ML applications in reservoir engineering, optimizing computational resources and improving predictive accuracy."

24 March 2025 12:45 till 13:35

[STAT/AP] Max Ramgraber: tba

25 March 2025 16:00 till 17:00

[AN] Cyril Houdayer: Noncommutative Poisson boundaries and Connes’ rigidity conjecture

Formulated in 1980, Connes' rigidity conjecture addresses a fundamental question in operator algebra theory: to what extent does the von Neumann algebra associated with a higher-rank arithmetic group preserve the structural information of the underlying algebraic group?
This talk will present a possible approach to this conjecture through the study of noncommutative Poisson boundaries of higher-rank arithmetic groups.
To make the presentation accessible, no prior technical knowledge in operator algebras or arithmetic group theory is required. The talk will include:
- An introduction to the construction of the von Neumann algebra of a discrete group
- An overview of Furstenberg's theory of Poisson boundaries
- A detailed application to the case of SL_n(ℤ) for n ≥ 3

31 March 2025 13:00 till 18:00

Afternoon on Scientific Machine Learning in Industry

Scientific Machine Learning (SciML) is rapidly transforming industries by integrating data-driven techniques with computational simulation models. While machine learning has revolutionized fields such as computer vision and natural language processing, its application in scientific computing introduces unique challenges in efficiency, accuracy, and robustness. The fusion of ML with numerical analysis and differential equations is opening new frontiers, addressing critical issues such as interpretability and the enforcement of physical constraints.

Industry is increasingly adopting SciML to enhance predictive modeling, optimize complex systems, and improve decision-making in fields such as engineering, finance, energy, and healthcare. Recent advancements, including physics-informed neural networks and neural surrogate models, are demonstrating SciML’s potential to accelerate innovation and scientific discovery. However, despite these successes, SciML methodologies still require extensive validation and robustness testing in real-world industrial environments, where complex interactions between machinery and environmental sensors pose unique challenges.

This event will bring together leading industry experts to explore the expanding role of SciML in industrial applications. The discussions will cover both opportunities and challenges, fostering a balanced dialogue on its impact. In addition to highlighting recent successes, the event will address risks associated with SciML deployment, including scalability concerns and the need for rigorous validation. The event will feature insightful talks from experts representing ASML, Robert Bosch GmbH, Deltares, and Siemens, along with ample opportunities for discussion and networking.

07 April 2025 12:45 till 13:35

[STAT/AP] Elisa Perrone: tba

10 April 2025 16:00 till 17:00

[PDE&A] Anh Khoa Doan: Scientific Machine Learning techniques for fluid mechanics

Machine learning techniques have in recent years seen an explosion in their development and application for fluid mechanics research problems. This is driven, in part, by the ability of machine learning approaches to model complex dynamics, allowing for improved predictions. Additionally, the embedding of physical information into deep learning architecture, to ensure the physicality of the predictions of the machine learning tools, is an on-going topic of research and has spurned architectures that can be used to solve inverse problems, and specifically augment the information content of a given dataset by reconstructing missing physical information.
In this talk, we will present machine learning applications along those two directions. First, we will present how a differentiable framework combined with deep learning can be used to complement an incomplete physical description of system. This will be illustrated on the case of a reacting flow, an archetypal multi-physics system that combines fluid mechanics and chemistry, the latter being the physics considered unknown. Second, we will present the use of physics-informed neural networks to augment experimental diagnostic. We will show how these can be used to reconstruct unmeasured quantities in a puffing pool fire and improve the spatial resolution of Particle Tracking Velocimetry measurements in a pulsed jet

28 April 2025 12:45 till 13:35

[STAT/AP] Alessandro Zocca

01 May 2025 16:00 till 17:00

[PDE&A] Gianne Derks

08 May 2025 16:00 till 17:00

[PDE&A] Rafael Bailo

12 May 2025 12:45 till 13:35

[STAT/AP] Taras Bodnar: Asymptotic properties of large dimensional Moore-Penrose and ridge-type inverses with applications

Large-dimensional asymptotic properties of the Moore-Penrose inverse and, as a byproduct, of various ridge-type inverses of the sample covariance matrix are derived. In particular, the analytical expressions of the asymptotic behavior of the weighted sample trace moments of generalized inverse matrices are deduced in terms of the partial exponential Bell polynomials which can be easily computed in practice. The existent results for the Moore-Penrose inverse are extended in several directions: (i) First, the population covariance matrix is not assumed to be a multiple of the identity matrix; (ii) Second, the assumption of normality is not used in the derivation; (iii) Third, the asymptotic results are derived under the large-dimensional asymptotic regime. Our findings provide universal methodology for construction of fully data-driven improved shrinkage estimators of the precision matrix, optimal portfolio weights and beyond. It is found that the Moore-Penrose inverse acts asymptotically as a certain regularizer of the true covariance matrix and it seems that its proper transformation (shrinkage) performs similarly to or even outperforms the existing benchmarks in many applications, while keeping the computational time as minimal as possible.

16 May 2025 12:30 till 13:15

[NA] Martin Campos-Pinto: TBA

TBA

22 May 2025 16:00 till 17:00

[PDE&A] Thomas Voigtmann

26 May 2025 12:45 till 12:13

[STAT/AP] Gokce Dayanikli: tba

02 June 2025 12:45 till 13:35

[STAT/AP] Tobias Muller: tba

16 June 2025 12:45 till 13:35

[STAT/AP] Antonis Papapantleon: tba

23 June 2025 12:45 till 13:35

[STAT/AP] Oliver Tse: tba

03 July 2025 16:00 till 17:00