Task Board

[hmomin]

1. Train LSTM agents on parallel time series environments to establish some baselines.
   1. PPO with LSTM (Assignee: Momin)
   2. TD3/SAC with LSTM (Assignee: Raghavan)
2. Construct parallel evolutionary training architecture with OpenAI ES on a single GPU for direct policy search. (Assignee: none)
3. Further scale evolutionary architecture from a single GPU to multiple GPUs using either a client-server model or a tournament-based ensemble design, as in ElegantRL-Podracer. (Assignee: none)

Once we've got all our training and environment architectures set up and working well, we can compare results across the deep reinforcement learning algorithms (PPO, TD3, SAC...) with those across evolutionary algorithms for various tasks.