Ulrik Unneberg

I'm a second year PhD student in Applied Mathematics in the Materials Intelligence Research lab at Harvard University supervised by Prof. Kozinsky. My current work is on using generative machine learning models to predict low probability quantum states.

I did my undergraduate studies in applied mathematics at the Norwegian University of Science and Technology, where I under supervision of Prof. Jakobsen wrote my thesis on numerically solving a set of nonlinear integro-differential equations, a revised version of which will soon be published. During my undergrad, I was fortunate enough to spend a year as a Fulbright fellow at MIT, where I worked with Prof. Lu on asymptotics of high-dimensional statistical theory.
Github     LinkedIn    

ulrikunneberg at gmail dot com

Data-Efficient Interatomic Potentials

Data-Efficient Interatomic Potentials by Fine-tuning NequIP and Allegro Foundation Models

Poster presentation to be held at Psi-K 2025, EPFL Switzerland.
Harvard University

Data-driven Multi-item Newsvendor Problem

The Data-driven Multi-item Newsvendor Problem: A Learning Theoretic Perspective

Learning theoretic analysis and numerical experiments on the data-driven multi-item newsvendor problem.
Harvard University
PDF  •   Code

Deep Reinforcement Learning for Memory-Driven Navigation

Deep Reinforcement Learning for Memory-Driven Navigation in Predator-Prey Environments

Investigating "memory" in predator-prey environments using deep reinforcement learning.
Harvard University
PDF  •   Code

Systolic Arrays for Efficient Attention Computation

Systolic Arrays for Efficient Attention Computation

We present a new architecture for computing the attention mechanism in the Transformer model using systolic arrays, where we fuse multilayer matrix operations.
Harvard University
PDF  •   Code

Fractional PDEs for Financial Modeling and Crowd Dynamics

A Numerical Method for Fractional Mean Field Games

Master's thesis on solving fractional partial differential equations with applications ranging from asset pricing to crowd dynamics.
Norwegian University of Science and Technology, Master's Thesis
PDF  •   Code

High-Dimensional Statistical Theory

Distribution of Score-Based Test Statistic for Gaussian Model Fails in the High-Dimensional Regime

Analyzing how theoretical results on M-estimators in the low-dimensional regime fails in the high-dimensional regime.
MIT, Fulbright Research
PDF  •  

2024

LondonML Intern at Palantir Technologies - Worked on a computer vision model for PDF parsing. Build out AI workflows for clients.

2023

OsloML Intern at Monil - Used varied machine learning methods (Ordinary NNs, CNN, LSTMs) to predict movement from time series.

2022

OsloAlgorithm Intern at Vind AI - Developed a new irregular optimizing algorithm of wind turbines, taking in wake effects, depth costs and cabling.

2024

Winnie & Ragnar Mathisen Prize - Ranked #1 student graduate across all engineering and architectural disciplines at NTNU.

2024

Stubban Prize - Ranked #1 in the mathematics graduates at NTNU.

2023

Aker Scholarship - Norway's #1 graduate scholarship.

2022

Fulbright Fellowship - Selected as one of six candidates for the non-degree Fulbright exchange fellowship, where I spent a year at MIT.