In the last blog we looked at the phenomenon of “jagged...
Read MoreCerenaut
Understanding animal intelligence
Improving machine intelligence
Research Strategy
We aim to discover new learning rules, architectures and representations from neuroscience and psychology to benefit AI and contribute insights back to these fields.
We are interested in the interactions of brain regions with complementary functions and timescales. For example, left and right hemispheres and slow and fast learning between neocortex and hippocampus.
Our focus is computational descriptions that are implementable.
FEATURED RESEARCH
ARROW: Augmented Replay for RObust World models
Inspired by evidence that hippocampal replay improves world knowledge, separate to policy learning, we added replay to a world model architecture for continual RL. A distribution-matching replay buffer within latent world models successfully prevents catastrophic forgetting with significantly reduced computational overhead.
Active perception and disentangled representations allow continual, episodic zero and few-shot learning
Generalization may not be required of every component in a learning system. We present a Complementary Learning System where the fast learner foregoes generalization in exchange for continual zero-shot and few-shot learning, operating as a parallel reasoner that biases the slow learner to encode novel stimuli in familiar, generalized terms.
NEWS 2026
- Jun 2026: ARROW is accepted into TMLR (see featured paper above)
- Mar 2026: Aziz authors ARROW, the WMAR extensions (see new featured research above)
- Feb 2026: Dave publishes a paper on active perception leveraging LLMs with episodic memory
- Feb 2026: Subbalakshmi starts as Research Engineer, on causal ML in medicine
- Feb 2026: Kevin begins a UNSW PhD on Executive Control (Model Based and Model Free RL)
- Jan 2026: David begins a Monash Masters on Executive Control (Model Based and Model Free RL)
- Jan 2026: Jingwen begins a Monash Masters extending L/R motor control
RECENT ARTICLES
Not like us: Why AI feels alien, and how to fix it
Introduction Will highly intelligent machines inevitably think like us? Our experiences...
Read More
HRM-Agent: Using the Hierarchical Reasoning Model in Reinforcement Learning
Authors: Hoang Long Dang and David RawlinsonUPDATE: Full HRM-Agent paper now...
Read More
Why I’m excited about the Hierarchical Reasoning Model
Since the arrival of Chat GPT, I’ve felt that progress...
Read More
Grant – Continual Few-Shot Learning
We are very happy to announce that we’ve been awarded...
Read MoreFEATURED ARTICLES
Video Prediction using Recurrent Sparse Memory
We recently presented 2 papers at the International Joint Conference...
Read More
One-shot learning for the long term: consolidation with an artificial hippocampal algorithm
We were really excited to present a couple of papers...
Read More
Towards Biologically Inspired Executive Control
Executive Control is core to what most people recognise as...
Read More
Sparse Unsupervised Capsules Generalize Better
We’ve just uploaded a spin-off research paper to arXiv titled...
Read MoreHow to build a General Intelligence: What we think we already know
Authors: D Rawlinson and G Kowadlo This is the first...
Read MoreWe co-supervise graduate projects
About Us
Cerenaut (formerly ProjectAGI) is an independent research group that undertakes fundamental research at the intersection of AI, Neuroscience and Psychology. We’re based in Australia.
Our name reflects humanity’s journey towards higher cognition and intelligence: cere = of the brain, naut = journey
We co-supervise student research projects and collaborate with researchers on topics of shared interest. If you’d like collaborate, get in touch!