Project AGI

Towards general-purpose machine learning


Algorithms for learning using only local and immediate credit assignment

High-order partially-observable sequences

Recurrent Sparse Memory (RSM) is an unsupervised method for simultaneously learning spatial and sequential structure in partially-observable high Markov order data streams

Fast (One-shot) episodic learning

Our Artificial Hippocampus Algorithm (AHA) can learn to generalize classes from a single instance, while still discriminating between instances of the same class - all without any provided labels

Research Strategy

We aim to discover new learning rules, architectures and representations by using computational neuroscience discoveries to guide our research


  • Oct 15 2019: Dave is presenting at the Melbourne MLAI meetup
  • Sep 2019: Episodic memory paper preprint available
  • July 2019: We welcome Bar Gal On as Commercial Director
  • June 2019: PhD student Shuai Sun begins APR internship with us to develop a clinical report feedback system in conjunction with Melbourne Health
  • May 2019: Recurrent Sparse Memory paper preprint available
  • Feb 2019: Satya Borgohain starts MSc project on self-organizing neural architectures co-supervised by us and Faculty of IT at Monash University


Project AGI is a privately funded research group based in Melbourne, Australia. We collaborate with academic institutions, companies and individuals on shared research interests. If you’d like to work with us, get in touch!


Incubator491 Pty Ltd is an Australian company that exists to facilitate research efforts such as Project AGI. Incubator491 also provides machine-learning consultancy services and intends to develop applied products in future. In particular, Incubator491 is interested in ML for healthcare.