Fundamental AI Research

Mirvum Research

Exploring the boundaries of intelligence.

Mirvum Research studies the foundations of intelligent systems: how models learn, how they remain stable, and how their uncertainty and decisions can be understood.

Explore our research Work with us
Stable structures An abstract geometric field representing robust structures emerging from uncertainty.
structure uncertainty stability

Why Mirvum

AI should not only perform.
It should hold up.

We focus on what remains stable when models are perturbed, shifted, retrained, or asked to explain themselves. Our goal is to develop methods that make intelligent systems more scientifically grounded, measurable, and dependable.

Research

Core directions

01

Model Stability

Understanding whether learned structures remain consistent across training runs, perturbations, and changing environments.

Foundations
02

Uncertainty & Verification

Measuring when models are uncertain, where their behavior becomes fragile, and how confidence can be validated.

Reliability
03

Interpretability

Developing explanations that are not only intuitive, but also stable, reproducible, and faithful to model behavior.

Understanding
04

Robust Intelligence

Studying how intelligent systems behave under distribution shift, noisy evidence, and real-world uncertainty.

Generalization

Principles

How we think about research

01

Start from first principles.

Build explanations and methods from measurable properties, not trends or slogans.

02

Stability is evidence.

When a learned structure persists under meaningful change, it becomes more scientifically informative.

03

Uncertainty should be visible.

Reliable systems should expose fragility rather than hide it behind a single confident output.

04

Research should transfer.

We look for ideas that generalize across models, data, domains, and generations of AI systems.

Publications

Research output

Selected papers, preprints, and technical notes will appear here.

Publication list coming soon.

Contact

Interested in fundamental AI research?

We are open to research collaborations, technical conversations, and partnerships around stable and reliable intelligent systems.

hello@mirvum.com Norway · International collaborations welcome