Felix Baastad Berg
I am an Aker Scholar pursuing an MSc in Advanced Computing at Imperial College London. My research interests are in reinforcement learning, high-performance computing, and statistical methods.
Previously, I was a Fulbright Scholar in Mathematics at MIT. I spent a year with the Rajan Lab at Harvard University, where I researched deep reinforcement learning simulations to model animal behavior.
I also hold a five-year integrated MSc in Mathematics from the Norwegian University of Science and Technology, where my thesis explored continuous attractor networks and memory in deep reinforcement learning.
Highlighted Research
Deep RL Needs Deep Behavior Analysis
We show model-free agents display implicit planning in open-ended environments and introduce a behavior-analysis toolkit to reveal how policies reason and generalize.
Paper (arXiv)Spatial Representation Learned in the Recurrent Memory of Artificial Neural Agents
Built biologically inspired reinforcement learning agents with LSTM memory, path integration, and continuous attractor networks. Designed analytical and numerical grid-cell modules, integrated into JAX-based neural networks. Statistical analyses of spatial encoding in recurrent memory.
Read Thesis (PDF)Rethought Generalization: Empirical Analysis of Bounds for Compositionally Sparse Networks
Conducted an empirical study of norm-based generalization bounds in CNNs under varying data sizes, random labels, and optimization strategies. Demonstrated when tighter bounds provide meaningful insight into generalization performance, and analyzed the effects of mini-batch size in SGD and regularization on generalization.
Unsupervised Time Series Forecasting with Spiking Neural Networks leveraging STDP
Implemented spiking neural networks with STDP for time series forecasting. Benchmarked performance against autoregressive models, evaluating predictive capacity on unclustered datasets.
Read Paper (PDF) CodeDeep Reinforcement Learning for Memory-Driven Navigation in Predator-Prey Environments
Developed a custom predator-prey grid world to evaluate memory in reinforcement learning agents. Found LSTM-based agents outperformed feedforward models and analyzed behavioral patterns including resource revisiting and predator evasion.
Professional Experience
Co-Founder (CTO → COO)
A digital learning platform for scientific courses with ~10,000 users. Led development of an interactive math-visualization library on top of Three.js, building innovative tools to enhance digital education.
ML Intern (Summer & Winter)
Worked as an LLM intern at a venture capital firm focused on scaling Norwegian technology companies. Built a LangChain-based chatbot that returned verified sources to support decision-making and organizational work.
Data Scientist (Intern and Part-time)
Created predictive models for rent prices in Python at Europe’s second-largest residential real estate company.
Data Science Intern
Built functionality and algorithms to analyze geospatial satellite data. Worked with Django, GeoPandas, and SQL to support renewable energy projects.
Lead of Developement of Simulator
Led development of a rocket flight simulator for a student-built rocket (Propulse NTNU) that ranked 2nd of 75 teams at Spaceport America Cup 30K COTS. Programmed in C++ and developed Computational Fluid Dynamics (CFD) simulations.
Honors
Aker Scholarship
Awarded Norway’s most prestigious graduate scholarship, providing full funding for advanced studies at leading global universities.
Fulbright Scholarship
Selected as one of six candidates for the non-degree Fulbright fellowship, where I spent a year at MIT.
Norwegian Finals of Physics Olympiad
Reached the national finals, ranking among Norway’s top high school physics students.