Real-time EEG decoding of selective attention for direct direct communication and control through thought – ComMental
The ComMental project focuses on decoding neural signals that reflect selective attention to specific contents, for direct mind control. It consists in identifying, within the EEG signal and in real-time, the characteristic features of the selective content the agent is paying attention to among a multitude of other contents. A robust (i.e. fast and precise) decoding of attention to that specific content, will allow, through a neurofeedback loop, for the corresponding command to occur. To do this, we will build on both the recent literature on brain-computer interface and machine learning, and on the know-how acquired within our laboratory regarding neuronal signatures of attention and the decoding of EEG signals for visual and/or auditory stimulation.
There is currently a real surge of interest in EEG, thank to recent technical advances that made the emergence of portable devices possible, not only for military and industrial sectors but also for large scale consumer applications. In particular, the miniaturization of electronic recording devices, and the important technical developments in the real-time decoding of cerebral responses, now allow for robust and ergonomic applications to be envisioned. These decoding techniques do not require the use of the classical evoked potential method, for which the response has to be locked to a given stimulus. Instead, they make use of the capacity to extract and use electrophysiological features for real-time prediction of the relationships between cerebral activity and perceptual contents.
To achieve a robust system for direct mind, the ComMental project will combine three main complementary research axes. The first axis will focus on decoding EEG in real-time for selecting visual contents and will mainly consist in refining the “proofs of concept” that we developed within the laboratory. The focus of the second axis will be the selection of auditory information and will consist in transposing the great performances that we obtained in the visual modality, this time for direct mind control of auditory contents, despite bigger constraints in the latter modality. The third axis will test different machine learning approaches, supervised and non-supervised, in order to develop learning models adapted to real-time EEG. To this end, we will compare some linear models traditionally used for EEG decoding of attention with more recent methods of non-linear patterns extraction using convolutional neuronal networks.
The ComMental project is hereby based both on fundamental research in cognitive neuroscience, and on applied research in brain-computer interface and machine learning. In a dual context, the methods developed here will have potential for both civil and military applications, either for achieving direct mind control when movement is limited, or for real-time monitoring of cerebral responses to surrounding stimuli.
Monsieur Sid KOUIDER (Ecole Normale Superieure)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
ENS Ecole Normale Superieure
Help of the ANR 298,338 euros
Beginning and duration of the scientific project: December 2017 - 30 Months