2019/08/06 by Nusrah Hussain, Engin Erzin, Hussain, Nusrah +5
Computer Science · Psychology · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Humor Studies and Applications #Machine Learning (cs.LG) #Multimodal Machine Learning Applications
paper · pdf · doi:10.48550/arxiv.1908.02037
openalex publication_date 2019/08/06 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
The ability to generate appropriate verbal and non-verbal backchannels by an\nagent during human-robot interaction greatly enhances the interaction\nexperience. Backchannels are particularly important in applications like\ntutoring and counseling, which require constant attention and engagement of the\nuser. We present here a method for training a robot for backchannel generation\nduring a human-robot interaction within the reinforcement learning (RL)\nframework, with the goal of maintaining high engagement level. Since online\nlearning by interaction with a human is highly time-consuming and impractical,\nwe take advantage of the recorded human-to-human dataset and approach our\nproblem as a batch reinforcement learning problem. The dataset is utilized as a\nbatch data acquired by some behavior policy. We perform experiments with laughs\nas a backchannel and train an agent with value-based techniques. In particular,\nwe demonstrate the effectiveness of recurrent layers in the approximate value\nfunction for this problem, that boosts the performance in partially observable\nenvironments. With off-policy policy evaluation, it is shown that the RL agents\nare expected to produce more engagement than an agent trained from imitation\nlearning.\n