BACS – Bayesian Approach to Cognitive Systems
Integrated Project conducted under the Thematic Priority: Information Society Technologies – Sub-topic: Cognitive Systems
Project duration: 01/01/2006 – 28/02/2010
Contract no: FP6-IST-027140
Project web page: http://www.bacs.ethz.ch/
- Autonomous Systems Lab, ETHZ
- CNBI, EPFL
- Max Planck Institute for Biol Cybernetics
- CNRS – LPPA, College de France
- FCT – Universidade de Coimbra
- INRIA – GRAVIR
- CNRS, Grenoble
- Hôpitaux Universitaires de Genève
- EDF (Electricité de France)
Our role in the BACS project was focused on EEG correlates of cognitive signals using Bayesian techniques. It is our aim to exploit real-time single trial recognition of these processes for a rich interaction with intelligent devices. This results in the development of semi-autonomous systems, in which an intelligent device (i.e., a Bayesian-based artificial cognitive system) is able to interact with a human user who provides corrective signals in order to improve the controller’s performance (i.e. Human in-the-loop). To attain this goal, we should be able to recognize EEG signals that convey useful information related to the system’s performance. Such information reflects user’s cognitive states such as error-recognition, anticipation of relevant events, alarm, as well as feedback related-psychophysical responses.
Error awareness in semi-autonomous systems
We focus the research on charaterizing error-related EEG potentials in interactive applications (Chavarriaga and Millán, 2010). Moreover, several experiments in conjunction with ETHZ, Zürich towards the design of artificial Bayesian controllers that reliably integrate Human-generated monitoring signals for navigation. This experiments include several user studies oriented to optimize the methods used to interpret the human monitoring signals, as well as the decisions made by the autonomous robot (Perrin et al., 2008, 2010). Experiments using multiple feedback modalities, as well as real and simulated robots were successfully performed.
Anticipation of future events
Following our work on characterization of anticipation-related potentials, we developed probabilistic techniques that achieve fast, reliable classification in single-trials. Furthermore, a first on-line implementation of an anticipation-based Brain-Computer interface was successfully implemented (Garipelli et al., 2008, 2009).
Decision making – Exploratory behavior
We study EEG correlates of exploratory behavior in decision making tasks. Following previous works using fMRI, we study human decision-making in a gambling task using several slot machines. In this task the subject’s decisions are taken either to obtain the highest payoff (i.e. exploitation) or to gather more information of the environment and improve future predictions (e.g. exploration).
We have shown that it is possible to discriminate between exploratory and exploitative behavior from EEG signals (Bourdaud et al., 2009). This was achieved by building behavioral models of decision making and novel classification algorithms for asynchronous EEG correlates. Furthermore, we explore new techniques for the analysis of these signals (Chavarriaga et al., 2008).