Ryan Julian

icon indicating a website link ryanjulian.me

icon company @google-deepmind, @rlworkgroup, @uscresl icon location San Franisco, CA icon location Research Scientist at Google DeepMind working on foundation models for real world world robots. Formerly at USC, Berkeley, Leap Motion, and Google X.

Results 39 issues of Ryan Julian

Add RNN support to torch/PPO and torch/TRPO

3 comment

feature
pytorch

Add GPU job to the CI

1 comment

tests
automation

Docs page "Usage Guide -> Automatic hyperparameter tuning"

9 comment

e.g. ray-tune and/or hyperopt

documentation

PyTorch on CPU is slower than TF

4 comment

See https://github.com/pytorch/pytorch/issues/975 for more info PyTorch TRPO appears 50% slower than TF. Not sure about PPO, but I expect the wall-clock time gap will be the same. To fix this...

pytorch

Ensure Python 3.8 support

1 comment

quality
tests
packaging

Docs page "Algorithms and Methods -> Relative Entropy Policy Search (REPS)"

documentation

Docs page "Algorithms and Methods -> CMA-ES"

documentation

Docs page "Algorithms and Methods -> Experience Replay -> PER"

documentation

torch/noisy DQN

See https://arxiv.org/abs/1710.02298

algos
pytorch

Docs Page "Algorithms and Methods -> Multi-Task RL -> TEPPO"

documentation