inference.mm

Matevz Matjasec

AI researcher · London, United Kingdom

If you’re working in a similar area, or you’re interested in my work, feel free to reach out.

About

I’m an AI researcher at the Max Planck Institute for Intelligent Systems, with a background in physics and computer science from Cambridge.

My work focuses on how optimisation, data, and architectural choices interact to shape learning dynamics, generalisation, and long-range behaviour in neural networks. I am particularly interested in the role of inductive biases in enabling models to move beyond pattern matching toward more structured, scientific reasoning.

Research interests

Inductive biases, architectures, and data in modern machine learning; learning dynamics and optimisation; generalisation and out-of-distribution robustness; in-context learning and long-context behaviour; diffusion-based models; natural language processing and genomic language modelling.

I am increasingly interested in the design of autonomous research systems—open-ended models capable of reasoning and knowledge generation beyond prompt-driven behaviour—and in how scalable, iterative research processes might be automated to accelerate scientific progress.

Current roles

Education

Interests

Walking in nature, skiing, sailing, hiking, jazz, opera, history, philosophy, politics.