Search papers, labs, and topics across Lattice.
Universit脿 degli Studi di Milano Politecnico di Milano, Universit脿 degli Studi di Milano
3
0
5
Adversarial bandits with $m$-set choices can suffer drastically slower learning rates than previously thought, with regret scaling quadratically with $m$ in some settings.
Log-barrier regularization can provably rescue policy optimization from getting stuck in suboptimal regions by structurally enforcing exploration, without sacrificing sample complexity.
The approximation error of spectral RL representations is fundamentally limited by the algebraic connectivity of the state-graph, revealing a crucial topological bottleneck.
