@gfranceschelli.bsky.social
2 followers 7 following 1 posts
Posts Media Videos Starter Packs
gfranceschelli.bsky.social
📢 New Paper!

We replace the entropy bonus in PPO with a *complexity* bonus, encouraging structured and stochastic policies that are robust to different scaling factors and can work in environments with variable exploration needs.

Read more:
arxiv.org/abs/2509.20509

w/ @mircomusolesi.bsky.social
Complexity-Driven Policy Optimization
Policy gradient methods often balance exploitation and exploration via entropy maximization. However, maximizing entropy pushes the policy towards a uniform random distribution, which represents an un...
arxiv.org