Theme

Machines that learn from interactions with people are becoming more common and more important parts of our lives. Interactive learning is regularly used in dialog systems, industrial robots, web search, and social media. It is pertinent to all forms of human-computer interaction, including medical robotics (e.g., bionic limbs, surgery systems, exoskeletons), the instruction of household appliances (e.g., thermostats, lights, fridges), and the dream of highly capable robot assistants. Learning from interaction is essential to any emerging technology that aims to amplify human cognitive, sensory, and physical capabilities through close partnerships between humans and intelligent machines.


Many systems that learn from humans have traditionally used supervised learning, but it is also natural, and becoming increasingly common, to use reinforcement learning methods. Whenever people interact with machines---be they prostheses, computers, mobile devices, or automobiles---there is a need to optimize the effectiveness of the interactions while minimizing the requirement for intrusive corrections or feedback. Reinforcement learning is natural for this setting because of the sequential and ongoing nature of the interaction, for which human feedback must be interpreted as evaluative, and because of the desire for the learning to continue for the long term, during the normal operation of the system (i.e., life-long, real-time learning). The general setting is that in which both the person and the machine are intelligent learning agents and perhaps are well-viewed as reinforcement learning systems. The idea of this workshop is to bring together experts in human-machine interaction, reinforcement learning, and the psychology of people learning from each other, to share ideas and explore possibilities for further research.


Potentially relevant sub-topics include (but are not limited to):

  • People interacting with robots

  • Dialog systems

  • Adaptive interfaces to computer systems

  • Intelligent adaptive prosthetic limbs

  • Brain-computer interfaces

  • Obtaining natural evaluative feedback from people

  • Learning from demonstration

  • Learning from social interaction

  • Observational Learning and imitation

  • Transfer learning from human to machine and vice versa

  • Transparency of learning systems & active learning

  • Collaboration and shared control

  • Communication between reinforcement learners