Policy Gradient Coagent Networks

By Philip S. Thomas. 2011

30 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Decoupled Neural Interfaces using Synthetic Gradients (2016, background): “…directly to each layer – e.g. REINFORCE (Williams, 1992) (considering all activations are actions), Kickback (Balduzzi et al., 2014a), and Policy Gradient Coagent Networks (Thomas, 2011) – but still remain update locked since they require rewards to be generated by an output (or a global critic).”
Routing Networks and the Challenges of Modular and Compositional Computation (2019, methodology): “Then the policy gradient theorems in [Thomas, 2011] for the acyclic and [Kostas et al., 2019] for the cyclic case hold, as the modules will also be updated with REINFORCE.”
Meta Learning Shared Hierarchies (2017, background): “Other past works (Thomas & Barto, 2011; Thomas, 2011; Thomas & Barto, 2012) have simultaneously learned modules that are used in conjunction to solve tasks, but do not incorporate temporal abstraction.”
On the Role of Weight Sharing During Deep Option Learning (2019, methodology): “Utilizing the coagent policy gradient theorems in Thomas (2011); Kostas et al. (2019), this mapping can then be used to dramatically simplify derivations of policy gradient theorems for complex policies.”
Learning Action Representations for Reinforcement Learning (2019, background): “To learn binary motor primitives, Thomas and Barto [2011] showed how a policy can be modeled as a composition of multiple “coagents”, each of which learns using only the local policy gradient information [Thomas, 2011].”