Avatar

Michiel Straat

Postdoctoral Research Group Leader “Life-long Machine Learning for Physical Systems”

Bielefeld University

Biography

I am a Postdoctoral Researcher and Junior Research Group Leader at Bielefeld University. Our Junior Research Group develops Artificial Intelligence methods for modeling and controlling complex physical systems. Motivated by major global challenges such as climate change and the energy transition, we explore how AI can help understand and model the dynamics of systems in fluid mechanics, and how to enforce desired outcomes through advanced control strategies. A key focus of our research is improving the robustness of these methods under changing conditions and increasing their sample efficiency by using surrogate models and embedding physical knowledge directly into the learning process.

Prior to this, I obtained my PhD (with Highest Distinction) from the University of Groningen, where I was part of the Intelligent Systems group. My doctoral research focused on the theory of neural networks, and I developed a successful predictive maintenance system for a smart factory environment.

Interests

  • Machine Learning for Physical Systems
  • Fluid Dynamics
  • Industry 4.0
  • Theory of Neural Networks

Education

  • Postdoc and Junior Research Group Leader

    Machine Learning group, Bielefeld University

  • PhD in Theoretical Machine Learning and Industry 4.0

    Intelligent Systems group, University of Groningen, PhD thesis

  • MSc. in Computing Science (specialization: Machine Learning)

    University of Groningen

Projects

SAIL

A multidisciplinary project focusing on transparent AI, human agency, safety and resource-efficiency.

Statistical Physics of Learning

Typical case analysis of machine learning scenarios

Reinforcement learning based control of dynamical systems

Employing AI techniques to model and control dynamical systems.

Machine Learning for health

Using Machine Learning methods to improve health care

Region of Smart Factories (RoSF), Industry 4.0

Focuses on the development of intelligent, connected and customized production processes

Recent Publications

Quickly discover relevant content by filtering publications.

Control of Rayleigh-Bénard Convection: Effectiveness of Reinforcement Learning in the Turbulent Regime

We build a robust reinforcement learning agent to control Rayleigh-Bénard Convection in turbulent conditions and compare against …

Solving Turbulent Rayleigh-Bénard Convection using Fourier Neural Operators

We train Fourier Neural Operator (FNO) surrogate models for Rayleigh-B’enard Convection (RBC), a model for convection processes …

Koopman-Based Surrogate Modelling of Turbulent Rayleigh-Bénard Convection

We build a linear dynamical model for complex fluid flows using a Machine Learning architecture based on Koopman theory.

Layered Neural Networks with GELU Activation, a Statistical Mechanics Analysis

In this work we show theoretically that using GELU activation in neural networks induces continuous phase transitions and we analyse …

Off-line Learning Analysis for Soft Committee Machines with GELU Activation

In this work we show that the GELU activation function in two-layer neural networks causes a continuous phase transition, independent …

An Industry 4.0 example: real-time quality control for steel-based mass production using Machine Learning on non-invasive sensor data

We present a typical Industry 4.0 case study: the real-time quality estimation of steel in a high-throughput production line using …

Supervised Learning in the Presence of Concept Drift: A modelling framework

A statistical physics based modelling framework is developed in which we study standard training algorithms (SGD, LVQ1) under concept …

Hidden Unit Specialization in Layered Neural Networks: ReLU vs. Sigmoidal Activation

Our systematic comparison of networks with ReLU and sigmoidal units in model situations reveals surprising differences in their …

Complex-valued embeddings of generic proximity data

We propose and evaluate an information-preserving complex-valued embedding for general non-psd proximity data.

On-line learning dynamics of ReLU neural networks using statistical physics techniques

We introduce exact macroscopic on-line learning dynamics of two-layer neural networks with ReLU units in the form of a system of …

Recent & Upcoming Talks

Solving Turbulent Rayleigh-Bénard Convection using Fourier Neural Operators

I present our publication on the use of Fourier Neural Operator models as surrogates for convection dynamics.

Solving Turbulent Rayleigh-Bénard Convection using Fourier Neural Operators

In this talk I present our work on harnessing zero-shot superresolution surrogate models FNO for convection.

Koopman-Based Surrogate Modelling of Turbulent Rayleigh-Bénard Convection

We presented our work on Koopman-based modeling of Rayleigh-Bénard Convection at the WCCI 2025 in Yokohama.

Koopman-based Modeling of Rayleigh-Bénard Convection

In this talk I will present our recent work on Koopman-based surrogate modeling of Rayleigh-Bénard Convection.

Statistical Physics of Learning

In this talk I discuss the main principles behind the statistical physics of learning.

Modelling adversarial training

In this talk I discuss my recent work on analyzing adversarial training procedures that make machine learning models more robust.

An Industry 4.0 example: real-time quality control for steel-based mass production using Machine Learning on non-invasive sensor data

In this talk a typical Industry 4.0 setting is presented: the use of sensor measurements for real-time quality control and fault …

Feedback Alignment methods for training neural networks

In order to compute weight updates, backpropagation uses complete knowledge of the downstream weights in the network. In [1] it is …

Dynamics of on-line learning in two-layer neural networks in the presence of concept drift

In this talk I will discuss our recent work on learning in non-stationary situations. A modelling scenario of neural networks learning …

Dynamics of on-line learning in two-layer neural networks in the presence of concept drift

Contact