Learning from human-generated reward

By W. B. Knox. 2012

60 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.

In Reinforcement Learning

Citations in the corpus

Deep Reinforcement Learning from Human Preferences (2017, background): “lear if the methods in Wilson et al. (2012) can be extended to complex tasks or if they can work with real human feedback. MacGlashan et al. (2017), Pilarski et al. (2011), Knox and Stone (2009), and Knox (2012) perform experiments involving reinforcement learning from actual human feedback, althoug”
Scalable agent alignment via reward modeling: a research direction (2018, background): “TAMER (Knox & Stone, 2009; Knox, 2012; Warnell et al., 2017) is a collection of algorithms that learn a policy from human value feedback, i.e. take actions that maximize expected feedback in the next step (possibly with short temporal smoothing).”
Interactive Learning from Policy-Dependent Human Feedback (2017, methodology): “Indeed, prior research has shown that interpreting human feedback like a reward function often induces positive reward cycles that leads to unintended behaviors [10, 11].”
Learning non-myopically from human-generated reward (2013, methodology): “Second, to account for delays in giving feedback, the causal attribution of each reward is distributed across multiple recent time steps by TAMER’s credit assignment module [5], further adding va-”
Training an Actor-Critic Reinforcement Learning Controller for Arm Movement Using Human-Generated Rewards (2017, background): “be possible [10], although careful parameter tuning was found to be essential to the success of such a technique [10].”
Learning Human Objectives by Evaluating Hypothetical Behavior (2019, methodology): “In this work, we align agent behavior with a user’s objectives by learning a model of the user’s reward function and training the agent via RL (Russell, 1998; Knox, 2012; Knox & Stone, 2015; Warnell et al., 2018; Leike et al., 2018).”
Improved human–robot team performance through cross-training, an approach inspired by human team training practices (2015, methodology): “In particular, SARSA( λ ) has been used to benchmark the TAMER framework (Knox and Stone, 2009), as well as to test TAMER-RL (Knox and Stone, 2010, 2012).”