Matteo
I explore how systems build useful world models, and how experience shapes what they learn. Here I share experiments in world modelling, curriculum, and self-supervision.
Playground
Try the research
Blogs
-
Amortised Assignment Generation
One step generation by efficiently pre-assigning gaussian latents to samples.
-
Curriculum Mining
Beating AlphaZero policy networks with search-free self-play PPO, diverse resets and sparse terminal rewards.
-
Pan1ni
Making a goal seeking policy with video pretraining only
-
SDMatch
Sketched Distribution Matching: a direct loss for matching one distribution to another with random or learned projections.
-
Self-Teaching Autoencoder
A self-training autoencoder. Trained without reconstruction losses, pixel or otherwise, and no image-space supervision of any kind.
-
Practical notes on leJEPA
Pitfalls to watch for whilst using leJEPA, and a couple of failed experiments along the way.
-
la leWorldModel
Simultaneous latent action and world model learning from passive video for controllable prediction.
-
Tokens, who needs them?
Improving transformer memory efficiency by 2.7x
-
Optimiser? I Hardly Know 'Er.
Using LLM Program Search to Build Robot Controllers Without RL Optimisers
-
Training a model to open the gates of hell
This is a short fun blog in which I showcase an experiment in creating endless learnable content.
-
Curriculum Is Key: Stepping Stones Toward AGI
Exploratory post on curriculum as the missing ingredient required for true intelligence and creativity.
-
Microbiome Modelling
Microbiome World Modelling.
-
Self-Organising Text
Cellular Automata Inspired Self Organising Text
-
Sample Efficient AR Diffusion
Oniris: Autoregressive and Sample-Efficient Next-Gen Video Diffusion
