kevin frans blog

Alchemist's Notes on Deep Learning

Small-Research: Tanh Activations with DDPG

Small-Research: Policy Extraction in IQL

Successor Representations Explained

Variational Information Bottleneck Explained

A Brief Examination of Generative Models

Deriving the KL divergence loss in variational autoencoders

For AGI, we need better tasks. For better tasks, we need open-endedness. (ALOE 2022 Notes)

A Mathematical Definition of Interestingness

To extract information from language models, optimize for causal response

Data digesters, ML^2, Interestingness

CLIPDraw: Exploring Text-to-Drawing Synthesis

StampCA: Growing Emoji with Conditional Neural Cellular Automata

Quality Diversity: Evolving Ocean Creatures

Open-Endedness 3: Multicell World