Collibra is a company that is specialised in data governance and that has developed a platform that organisations can use to manage their collection of data assets (reports, data sets, analyses, etcetera). The project aims to develop new ways in which members of such organisations in various roles can engage with their data. In particular, we explore the use of chatbots operating on messaging platforms such as Slack, allowing users to interact with each other as well as with automated apps using natural language. At the VUB AI Lab, we are interested in developing interactive machine learning techniques, allowing data scientists to engage with data sets, using their domain expertise to potentially improve the models that were learned from the data.
The Multi-Dimensional Interactive management and Adaptive Learning project aims to combine the linguistic theory of multi-dimensionality in dialogue (Bunt, 2011; Keizer & Bunt, 2006, 2007) with statistical approaches to dialogue management (Young et al, 2010, 2013), in order to develop dialogue systems that 1) support more natural and socially-intelligent conversations, and 2) enable more efficient cross-domain adaptation by exploiting the separation of domain-independent and domain-general aspects of communication. In particular, we have developed a dialogue manager that uses multi-agent reinforcement learning to train action selection policies for each dimension, allowing transfer and adaptation of the domain-independent policies to a new domain.
Strategic Conversation. In this project, we focused on non-cooperative dialogue, and developed reinforcement learning strategies for trade negotiation in the context of the board game Settlers of Catan. [website]
Joint Action for Multimodal Embodied Social Systems. This was a project on human-machine interaction, social robotics, and computer vision, in which we developed a proof-of-concept robot bartender. At Heriot-Watt University, we focused on machine learning techniques for speech-enabled multi-user human-robot interaction. [website]
Computational Learning in Adaptive Systems for Spoken Conversation. [website]
An EPSRC funded project, focusing mainly on the Hidden Information State (HIS) POMDP dialogue system.
PARallel Agent-based DIalogue Management Engine, project with Harry Bunt at Tilburg University on multi-dimensional dialogue management in the context of interactive question-answering. [IMIX book]