Beren's Projects
A predictive processing library/framework I was working on a while ago.
Some quick experiments with Feedback Alignment and Direct Feedback Alignment
Experiments with learning transition matrices for LQR by gradient descent by differentiating through them with Zygote
Replicating and dissecting the git-re-basin project in one-click-replication Colabs
Experiments with Divisive Normalization of prediction errors in predictive coding as in Spratling
The mozilla tutorial and others as I try to get to grips with django
My reimplementation of Manuel Baltieri's active inference agent (Ch6 of his thesis)
Dream to Control: Learning Behaviors by Latent Imagination
Experiments with AR with dynamic weights and other relaxations to the algorithm
Experiments with dynamical predictive coding -- i.e. no fixed predictions, continual weight updates etc.
Some messing around with evolutionary game theory models
Some quick work trying to learn about and implement evolutionary algorithms
A set of examples around pytorch in Vision, Text, Reinforcement Learning, etc.
Baselines and environments for exploration and model-based reinforcement learning
A quick parser for the eyelink xml files containing word stimulus information
A custom parser for parsing eyelink ASC files into csv files.
Some infrastructure to simulate more biologically realistic visual inputs and eye movements
Code for various eye tracking analyses I did during my PhD. Primarily in R for LMER modelling but heavy duty computation in julia or python
Experiments with controlling computer mouse with gaze and head gestures
A repository for major/influential FEP and active inference papers.
Just some misc doodles
Experiments with fingerprinting LLMs based on unconditioned distribution
Relax! Flux is the ML library that doesn't make you tensor
Some stuff related to game theory I think is interesting
Gestalt autoencoders using a generative Variational Autoencoder Model.
Replication of Generative Control for Free paper... and it works!!!
Some brief phd work on gestalt autoencoders which learn to predict one half of an image from the other
Some fun stuff with GPs