Master thesis: Bridging Physics and Machine Learning through Explainable Image Generation

1 day ago

Taby Sweden Mycronic Full time 540,000 - 840,000 per year
Description

Bridging Physics and Machine Learning through Explainable Image Generation

Background and Motivation
In advanced industrial applications, physics-based simulation models are widely trusted for their interpretability and strong theoretical foundations. However, these models often struggle to describe the wide range of complex and nonlinear effects that occur in real-world situations, such as optical distortions, displacement, motion blur, and noise.

Machine learning (ML) models, particularly generative approaches capable of synthesizing or reconstructing images, offer powerful alternatives that can learn such relationships directly from data. Yet, skepticism remains high among experts, mainly due to concerns about reliability, stability, and explainability.

This thesis aims to bridge that gap by combining Explainable AI (XAI) and Physics-Informed Neural Network (PINN) principles. By integrating physical constraints and interpretability mechanisms into ML-based image generation, the project seeks to make these models transparent, interpretable, and consistent with known physical behavior.

Objectives

  • Develop an explainable image generator that visualizes and quantifies how physical phenomena are represented within the model.
  • Integrate physics-inspired loss functions or regularization terms (e.g., illumination smoothness, diffraction consistency) based on PINN concepts to ensure the network respects known physical relationships.
  • Evaluate model robustness and reliability under controlled perturbations such as defocus, brightness variation, alignment shifts, and unexpected noise.
  • Produce visual and numerical explainability results that correlate ML reasoning with physical intuition — strengthening both understanding and trust.

Expected Impact

  • Scientific understanding: Reveal how ML models implicitly capture or deviate from physical laws.
  • Industrial confidence: Demonstrate transparent and reliable ML-based models that can be trusted alongside physics-based methods.
  • Long-term value: Establish a foundation for hybrid ML–physics modeling, enabling interpretable AI systems.

Candidate Requirements

  • Solid understanding of image processing and optical or physical modeling principles.
  • Strong background in machine learning and computer vision (TensorFlow or PyTorch and OpenCV).
  • Experience with data-driven modeling, including CNNs, GANs, UNets, and other generative architectures.
  • Interest in explainable AI, model interpretability, and hybrid ML–physics integration.

Preferred Skills

  • Familiarity with Fourier optics or physics-informed neural networks (PINNs).
  • Experience with model visualization, attribution techniques, or uncertainty quantification.
  • Ability to communicate technical insights effectively to both ML and physics-focused audiences. 

  • Master thesis: Active learning for deep-learning-based deviation classification in photomasks

    1 day ago

    Taby, Sweden Mycronic Full time 60,000 - 120,000 per year

    DescriptionAbout MycronicMycronic is a global high-tech company whose innovative solutions have been advancing electronicstechnology for over 40 years. Our innovative solutions, including our market-leading mask writers, have madea significant impact on the electronics industry. Today we continue to grow and serve customers in anexpanding variety of...

  • Master Thesis

    3 days ago

    Sweden Atlas Copco Full time 350,000 - 550,000 per year

    Your role Level of thesis project: Master thesisEducation requirement: Computer Science, Artificial Intelligence, Machine Learning, or a related fieldFunctional area: Atlas Copco Industrial Technique AB, Application SoftwareNumber of students for this project: 1Start: Flexible, early 2026 preferredWe are seeking a curious and highly driven Master student...

  • Master Thesis: Finite Element and Machine Learning for Dynamic Cable Sensors

    3 days ago

    Sweden NKT Full time 60,000 - 90,000 per year

    BackgroundMotivated by the urgent need to address climate change, the generation of renewable energy is on the rise. To reach out for new natural resources, such as wind, wind farms are now more often being installed offshore at sea, further off the coast. In such places, winds blow with much greater force and more frequently than on land. High voltage (HV)...

  • Machine Learning Engineer

    1 day ago

    Östra Hamngatan Göteborg, Sweden, Sweden Modulai Full time 80,000 - 120,000 per year

    MACHINE LEARNING ENGINEERAs a member of the ML-team you will be working with a broad range of problems with one common denominator – ML will be the key ingredient.You will have to analyze the problem at hand, come up with a solution strategy and execute it. This typically entails gaining an in-depth understanding of the challenge, understanding the...

  • Senior Machine Learning Engineer

    1 day ago

    Östra Hamngatan Göteborg, Sweden, Sweden Modulai Full time 120,000 - 240,000 per year

    SENIOR MACHINE LEARNING ENGINEER (SMLE) - SwedenAs a member of Modulai you will be working with a broad range of problems with one common denominator – ML will be the key ingredient. As a SMLE you will be responsible for team success and advisory to both teams and clients. In this role, you will be responsible for engaging clients and teams from the...

  • Master Thesis Application for Spring 2026

    3 days ago

    Typsnittgatan Uppsala, Sweden Nodica Group AB Full time 40,000 - 80,000 per year

    Who We AreScandiNova develops high-technology pulse modulators used in medical technology, research, and industry – including applications such as cancer therapy, particle accelerators, and weather radar systems.Master's Thesis at ScandiNovaAt ScandiNova, we highly value thesis projects and the contributions of our students. We are therefore always eager...

  • Examensarbete. Correlation and optimization of theoretical 3D laser cutting simulation

    1 day ago

    Ektjärnsvägen Luleå, Norrbotten, Sweden Gestamp HardTech AB Full time 40,000 - 80,000 per year

    Master Thesis at Gestamp HardTech. Department: TechCenterCorrelation and optimization of theoretical 3D laser cutting simulationGestamp HardTech in Luleå is part of the global Gestamp group which develops and manufactures components for the automotive industry. Gestamp HardTech serves many well-known customers and specializes in the production of...

  • PhD Candidate Quantum

    3 days ago

    Sweden Atlas Copco Full time 4,504,000 - 7,206,000 per year

    Your role Your mission is to advance research in quantum technologies by joining the Wallenberg Quantum Centre as a PhD candidate registered at Chalmers University of Technology (Sweden) while being employed by Atlas Copco Group on a 4-year contract. This role is part of Atlas Copco's Scientific Vacuum Division quantum strategy, focusing on...

  • Machine Learning Engineer

    3 days ago

    Bollgatan Växjö, Växjö, Sweden Fortnox AB Full time 900,000 - 1,200,000 per year

    Vi tror på ett samhälle format av framgångsrika företag där entreprenörskap och samhällsnytta går hand i hand. Att vara en del av Fortnox är att vara en del av en ständig förändring - en förändring där du själv bestämmer hur långt du vill nå. Vårt största fokus är att ge dig rätt förutsättningar att få leva din passion tillsammans...

  • Machine Learning Engineer

    1 day ago

    Bollgatan Växjö, Växjö, Sweden Fortnox AB Full time 450,000 - 900,000 per year

    Vi tror på ett samhälle format av framgångsrika företag där entreprenörskap och samhällsnytta går hand i hand. Att vara en del av Fortnox är att vara en del av en ständig förändring - en förändring där du själv bestämmer hur långt du vill nå. Vårt största fokus är att ge dig rätt förutsättningar att få leva din passion tillsammans...