We survey ideas from complex systems such as swarm intelligence, self-organization, and emergent behavior that are gaining traction in ML. (Figure: Emergence of encirclement tactics in MAgent.)
Introduction
Unless you’ve been living under a rock…
EvoJAX is a hardware-accelerated neuroevolution toolkit built on top of JAX. It can help run a wide range of evolution experiments within minutes on a TPU/GPU, compared to hours or days on CPU clusters.
Redirecting to github.com/google/evojax/, where the repo resides.
Reinforcement learning agents typically perform poorly if provided with inputs that were not clearly defined in training. A new approach enables RL agents to perform well, even when subject to corrupt, incomplete, or shuffled inputs.
Note: This…
<!–
–>
<!–
–>
<!–
–>
“A drawing of a cat”
CLIP + ES + Triangles
<!–
GitHub
Redirecting to es-clip.github.io, where the article resides.
<!–
–>
<!–
–>
<!–
–>
<!–Evolved Biped Walker.
–>
Agents with a self-attention “bottleneck” not only can solve these tasks from pixel inputs with only 4000 parameters, but they are also better at generalization.
<!–
GitHub
Redirecting to attentionagent.github.io, where the article resides.
<!–
–>
<!–
–>
<!–
–>
<!–Evolved Biped Walker.
–>
Rather than hardcoding forward prediction, we try to get agents to learn that they need to predict the future.
<!–
GitHub
Redirecting to learningtopredict.github.io, where the article resides.
<!–
–>
<!–
–>
<!–
–>
<!–Evolved Biped Walker.
–>
We search for neural network architectures that can already perform various tasks even when they use random weight values.
<!–
GitHub
Redirecting to weightagnostic.github.io, where the article resides.
<!–
–>
<!–
–>
<!–
–>
<!–Evolved Biped Walker.
–>
PlaNet learns a world model from image inputs only and successfully leverages it for planning in latent space.
<!–
GitHub
Redirecting to planetrl.github.io, where the article resides.
<!–
–>
<!–
–>
<!–
–>
<!–Evolved Biped Walker.
–>
Little dude rewarded for having little legs.
<!–
GitHub
Redirecting to designrl.github.io, where the article resides.
GitHub
In this article I will give step-by-step instructions for reproducing the experiments in the World Models article (pdf). The reference TensorFlow implementation is on GitHub.
Other people have implemented World Models independently. The…