Search Results for author:
Found 4 papers, 1 papers with code
Behind the Myth of Exploration in Policy Gradients
To compute near-optimal policies, it is essential in practice to include exploration terms in the learning objective.
Informed POMDP: Leveraging Additional Information in Model-Based RL
We then show that this informed objective consists of learning an environment model from which we can sample latent trajectories.
Recurrent networks, hidden states and beliefs in partially observable environments
In summary, this work shows that in its hidden states, a recurrent neural network approximating the Q-function of a partially observable environment reproduces a sufficient statistic from the history that is correlated to the relevant part of the belief for taking optimal actions.
Warming up recurrent neural networks to maximise reachable multistability greatly improves learning
This insight leads to the design of a novel way to initialise any recurrent cell connectivity through a procedure called "warmup" to improve its capability to learn arbitrarily long time dependencies.
