AI systems are becoming an increasing everyday presence for the average person, but how the systems should interact with humans and how they could be optimal at supporting users with their tasks is often debated.
Main research project
Effective human-AI interaction
The group’s overall research goal is to create methods and algorithms for more effective human-AI interaction. To achieve this goal, we employ human behavioral data, which is becoming more and more available due to the increased network connectivity of many everyday devices. This data is used to train and improve the AI systems and generate predictive models of human behavior that enable the AI to better adapt to its users. In addition to this, we study the use of reinforcement learning to train intelligent assistants that can be used to support the work of human users.
In addition to tackling fundamental research questions on human-AI interaction, we are targeting a wide range of application areas in our work, including:
- Software Development, e.g. intelligent programmer assistance.
- MOOCs, e.g. learner performance prediction and personalized exercise generation.
- Robotics, e.g. socially aware robots.
Continuous Gesture Recognition from sEMG Sensor Data with Recurrent Neural Networks and Adversarial Domain AdaptationICARCV,
Learning to Run with Reward Shaping from Video DataWorkshop on Adaptive and Learning Agents (ALA) at ICML-AAMAS,
Introspective Reinforcement Learning and Learning from Demonstration17th International Conference on Autonomous Agents and Multiagent Systems (AAMAS),
Learning to Run with Potential-Based Reward Shaping and Demonstrations from Video DataICARCV,