Skip to content
Selma Holmen Berg

Applied Physics & Mathematics · NTNU

Selma Holmen Berg

Machine Learning Engineer at Women in AI × Cognite

Third-year student in Applied Physics and Mathematics at NTNU, specialising in Industrial Mathematics. I work where mathematics meets real data, from neural networks that respect physics to machine learning on large industrial datasets.

Machine learning · Neural networks · Agentic AI · Graph analytics · Scientific computing · Numerical modelling · Data analysis · Optimisation · Python

Portrait of Selma Holmen Berg, smiling.

Selected work

Projects

  1. Link Prediction and Outlier Detection in Industrial Knowledge Graphs

    • Women in AI × Cognite
    • NTNU Women in AI programme
    • Ongoing
    • Team project

    Using machine learning to find errors and missing connections in an industrial knowledge graph, together with Cognite.

    Diagram of a knowledge graph: maintenance orders (M1–M3) linked to operations, which link to equipment, time series and assets. Some nodes are isolated.
    • Machine learning
    • Graph embeddings
    • Node2Vec
    • Outlier detection
    • Link prediction
    • Knowledge graphs
    • Data analysis
    • Python

    Sep 2026 – presentRead more

  2. Function Approximation: Interpolation, RBFs and Optimised Nodes

    • NTNU
    • TMA4215 Numerical Mathematics
    • Individual

    Comparing classic ways to estimate a function between known points, finding out where they fail, and optimising where to place the points.

    Runge's function interpolated with 11 equidistant nodes, which oscillate strongly near the ends, and with 11 Chebyshev nodes, which stay close to the function.
    • Numerical analysis
    • Interpolation
    • Optimisation
    • Gradient descent
    • Automatic differentiation
    • Python

    Sep 2026Read more

  3. Physics-Informed Neural Networks for Indoor Temperature

    • NTNU
    • TMA4320 Introduction to Scientific Computing
    • Group of 3
    • Grade A

    Estimating the temperature everywhere in a room from just nine sensors, by teaching a neural network the physics of heat.

    Diagram of a fully connected neural network with inputs x, y and t, three hidden layers and output temperature T. Figure from the TMA4320 assignment text.
    Fully connected network: (x, y, t) → T
    • Finite-difference solution after 24 hours: a round warm area around the heat source.Numerical solution
    • Neural network prediction after 24 hours: a diagonal warm band across the room.Neural network
    • PINN prediction after 24 hours: a smooth warm area around the heat source.PINN
    • Machine learning
    • Neural networks
    • Physics-informed ML
    • Deep learning
    • JAX
    • Automatic differentiation
    • PDEs
    • Python

    Feb 2026Read more

  4. Stochastic Simulation of Interacting Motor Proteins

    • NTNU
    • TMA4320 Introduction to Scientific Computing
    • Group of 3
    • Grade A

    A computer simulation of how proteins inside cells carry cargo in one direction, even though every step they take is random.

    Positions of 20 interacting particles over time: each line moves in small jumps and stays flat while blocked by a neighbour.
    • Monte Carlo simulation
    • Stochastic modelling
    • Statistical physics
    • Scientific computing
    • Python

    Mar 2026Read more

  5. Numerical Boundary-Value Problems and the Josephson Effect

    • NTNU
    • TMA4320 Introduction to Scientific Computing
    • Group of 3
    • Grade A

    Writing our own equation solvers, then using numerical methods to compute the current through a superconducting junction.

    Density of states in the middle of the junction against energy for three lengths: a gap near zero energy that narrows as the length grows.
    • Numerical methods
    • ODE solvers
    • Boundary-value problems
    • Scientific computing
    • SciPy
    • Python

    Apr 2026Read more

Beyond the code

Hobbies

Poker club · Leader, fall 2025 – fall 2026

Nabla Casinus

Ran the weekly poker nights for Nabla, the student association for physics and mathematics at NTNU. Attendance grew from 6 to over 50 students.

Kiting · since I was 15

Tougher than I look

I started kiting when I was 15, and I have kited through storms on Hardangervidda with winds of 20+ m/s.

Cooking · Cookbook

Smakfulle, sunne og enkle middager – for enkle folk

I love cooking, and making good food feels like the best way to unwind after a long day. I wrote, designed and gave my parents a cookbook: simple recipes, quality ingredients and clear instructions for slightly chaotic, impatient people. Made with Photoshop and PowerPoint.