3 ms·
PPO is awesome, but so is GPT-style reward-trajectory prediction! http://arxiv.org/pdf/2106.01345v1 http://arxiv.org/pdf/2106.01345v1. As a RL hobbyist, I'd lo
by isaacimagine 5y ago
PPO is awesome, but so is GPT-style reward-trajectory prediction! http://arxiv.org/pdf/2106.01345v1 http://arxiv.org/pdf/2106.01345v1.
As a RL hobbyist, I'd love to see some sort of hybrid approach. Thoughts?