Search papers, labs, and topics across Lattice.
Affiliation:
3
0
2
Discontinuous boundary dynamics no longer prevent finite-time guarantees in two-time-scale RL, unlocking rigorous $O(n^{-1}\log n)$ value-gap rates for projected actor-critic under dependent Markovian noise.
Residual connections can dramatically reshape the Liapunov spectrum, offering a pathway to stabilize gradient behavior in deep networks.
Q-learning regret bounds can be achieved without optimism, but are highly sensitive to the suboptimality gap, motivating a new smoothed exploration strategy.
