Timezone: »

Learning to Generalize from Sparse and Underspecified Rewards
Rishabh Agarwal · Chen Liang · Dale Schuurmans · Mohammad Norouzi

Thu Jun 13 06:30 PM -- 09:00 PM (PDT) @ Pacific Ballroom #49

We consider the problem of learning from sparse and underspecified rewards, where an agent receives a complex input, such as a natural language instruction, and needs to generate a complex response, such as an action sequence, while only receiving binary success-failure feedback. Such success-failure rewards are often underspecified: they do not distinguish between purposeful and accidental success. Generalization from underspecified rewards hinges on discounting spurious trajectories that attain accidental success, while learning from sparse feedback requires effective exploration. We address exploration by using a mode covering direction of KL divergence to collect a diverse set of successful trajectories, followed by a mode seeking KL divergence to train a robust policy. We propose Meta Reward Learning (MeRL) to construct an auxiliary reward function that provides more refined feedback for learning. The parameters of the auxiliary reward function are optimized with respect to the validation performance of a trained policy. The MeRL approach outperforms an alternative method for reward learning based on Bayesian Optimization, and achieves the state-of-the-art on weakly-supervised semantic parsing. It improves previous work by 1.2% and 2.4% on WikiTableQuestions and WikiSQL datasets respectively.

Author Information

Rishabh Agarwal (Google Research, Brain Team)
Chen Liang (Google Brain)
Dale Schuurmans (Google / University of Alberta)
Mohammad Norouzi (Google Brain)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors