Explore projects

This repo provides the necessary source code to simulate itinerant dynamics, that is, dynamics that randomly switch between attractor patterns, using a RNN implementation based on predictive coding.
Updated 
Predictive coding algorithm for memory retrieval in RNNs. Based on a target sequence, we can infer the latent discrete representation (or class) that best predicts it.
Updated 
A model using SOMs and active inference to learn a repertoire of simple motor trajectories
Updated 
Example of learning, prediction and inference using a model of RNN based on predictive coding.
Updated 
Example implementation of a twolayered network with predictive codingbased inference and learning.
Updated 
A benchmark of online learning algorithms for RNNs in a continual learning setting. The models are evaluated on the task of incrementally learning a repertoire of 20 continuous sequences.
Updated 

Updated