Ongoing MuZero and Cognitive Planning Investigate the relation between the famous MuZero algorithm and human cognitive planning. ET & IT neuron modelling Building a computational model of cortical ET and IT pyramidal neurons during learning of a stimulus-reward association task. Action Gradients in the Brain Propose a 'action gradient' framework that may help explain Cerebellar and Basal Granglia interactions during learning. Past Taylor TD-learning Provides lower variance TD updates through a first-order Taylor expansion of expected TD updates. MB-DPG Provides a RL policy gradient algorithm that can capture human motor adaptation under visuomotor perturbations. EpiQ Attempts to estimate epistemic uncertainty of value functions, by approximating a distribution over transition models as well as value functions through model ensambles. Extra Protein design with LLMs Fine-tune a pretrained protein-based generative large language model on a specific dataset, using supervised fine-tuning and DPO. Transformer from scratch My own implementation of an encoder-decoder Transformer, trained on a random sequence dataset. Popular deep RL algorithms My own implementation of the most popular deep RL algorithms for both discrete and continous action spaces.
