Mikel Bober-Irizar

Mikel Bober-Irizar

a.k.a. anokas / mxbi

Hey!

I’m a machine learning engineer and researcher based in London. I like building at the intersection between AI and other fields, and have previously worked on hardware, privacy, security, ancient Japanese literature and more. I'm best known for my work on Kaggle where I was the youngest ever competitions grandmaster (and #1 in the UK overall).

I currently work as a Founding Principal Engineer at Iconic Games, where I lead AI development on our first game, as well as contributing to game design and development. I’m excited about the role that runtime ML models can play in building entirely new experiences and genres of games, and can’t wait to share this project with the world.

I continue to do research as a visitor at the University of Cambridge (and occasionally the University of Surrey), where I graduated with an MEng in 2023. My main research focus is currently on machine reasoning.

I’m also a (very part-time) Senior Machine Learning Scientist at ForecomAI, advising on bioscience research.

I’d love to chat, the best way to contact me is usually over email. I'm always interested in hearing from people with similar interests.

Cat animation.

Games

I’ve always been passionate about video games as a medium and the way that it brings together so many technical and creative disciplines. I’m now fortunate to work with artists, writers and designers every day - but still occasionally like to participate in game jams with my friends.
Here are some of those games; let me know what you think!

Game 4

Project ███

Iconic's first game. I can't tell you much except that it's in space and it's like nothing you've played before. Coming 2025?
Game 1

Rock Paper Summon Frog Plutonium • 2024

Ludum Dare Jam 55.
You are among the select few to realize your fantasies of vanquishing a Lemon with an Air Fryer.
Game 2

Beaneth the Earth • 2023

Cambridge Game Jam '23.
✨ won Best Gameplay
A tower defence game where surviving is really easy and winning is really hard.
Game 3

Goblin Heist • 2021

Ludum Dare Jam 54.
A short and very speedrunnable 2D puzzle game.

Research

I've done research in all areas of applied AI, affiliated with various organisations. Here's some of my favourite work from the last couple years, mostly with the University of Cambridge, and some tl;dr summaries.

LLMs struggle with perception, not reasoning, in ARC-AGI 2024
Reasoning LLMs

A deep dive into o3's stellar performance on the ARC-AGI benchmark, what makes it hard for LLMs, and what benchmark performance is really measuring.

Neural networks for abstraction and reasoning 2024
Reasoning LLMs

Exploring two new approaches to the then-stalled ARC-AGI benchmark, based on neurosymbolic reasoning and LLMs.

Skill Issues: An Analysis of CS:GO Skill Rating Systems 2024
Games

An empirical analysis of the performance ofElo, Glicko2 and TrueSkill skill rating systems in Counter-Strike: Global Offensive, based on surrogate modelling of professional matches.

Single-cell subcellular protein localisation using novel ensembles of diverse deep architectures 2023
Vision Healthcare

Developed ensemble deep learning approaches for protein localization in cellular imaging, achieving state-of-the-art accuracy.

Architectural Backdoors in Neural Networks 2022
Security

Introduced Model Architecture Backdoors, where an exploitable deficiency can be hidden directly in the model architecture and persist through retraining.

See all research (14)

Contact

My email is [email protected], or send me a message on Twitter if you want to work together or chat!
I may be open to consulting work, but I'm not looking for full-time roles right now.