Learning From EEG Error-Related Potentials in Noninvasive Brain-Computer Interfaces

Abstract

Abstract-We describe error-related potentials generated while a human user monitors the performance of an external agent and discuss their use for a new type of brain-computer interaction. In this approach, single trial detection of error-related electroencephalography (EEG) potentials is used to infer the optimal agent behavior by decreasing the probability of agent decisions that elicited such potentials. Contrasting with traditional approaches, the user acts as a critic of an external autonomous system instead of continuously generating control commands. This sets a cognitive monitoring loop where the human directly provides information about the overall system performance that, in turn, can be used for its improvement. We show that it is possible to recognize erroneous and correct agent decisions from EEG (average recognition rates of 75.8% and 63.2%, respectively), and that the elicited signals are stable over long periods of time (from 50 to 600 days).Moreover, these performances allow to infer the optimal behavior of a simple agent in a brain-computer interaction paradigm after a few trials.Index Terms-Brain-computer interface, electroencephalography (EEG), error-related potentials, reinforcement learning.

Brain connectivity in continuous error tasks

“…Signal was standardized for every sliding window (subtracting temporal mean and dividing by the standard deviation) and the difference between error and correct coherence was computed for the most significantly different pairs of channels. For comparison purposes, the grand averaged signals were computed for the error and correct conditions at channel FCz [8]. Event-related synchronizations (ERS) were calculated in a time window of -1500 to 1500 ms using Morlet wavelets, with a wavelet-width of 12 cycles.…”

“…A statistical significance test based on bootstrapping was run over the ERSs following the method described in [20]. O2 P8 P4 CP6 T8 C4 FC6 F8 F4 AF4 Fp2 Oz Pz CP2 CPz CP1 Cz FC2 FCz FC1 Fz O1 P7 P3 CP5 T7 C3 FC5 F7 F3 AF3 P8 P4 CP6 T8 C4 FC6 F8 F4 AF4 Fp2 Oz Pz CP2 CPz CP1 Cz FC2 FCz FC1 Fz O1 P7 P3 CP5 T7 C3 FC5 F7 F3 AF3 Fp1 alpha band (8)(9)(10)(11)(12) Fz O1 P7 P3 CP5 T7 C3 FC5 F7 F3 AF3 O2 P8 P4 CP6 T8 C4 FC6 F8 F4 AF4 Fp2 Oz Pz CP2 CPz CP1 Cz FC2 FCz FC1 Fz O1 P7 P3 CP5 T7 C3 FC5 F7 F3 AF3 P8 P4 CP6 T8 C4 FC6 F8 F4 AF4 Fp2 Oz Pz CP2 CPz CP1 Cz FC2 FCz FC1 Fz O1 P7 P3 CP5 T7 C3 FC5 F7 F3 AF3 Fp1 P8 P4 CP6 T8 C4 FC6 F8 F4 AF4 Fp2 Oz Pz CP2 CPz CP1 Cz FC2 FCz FC1 Fz O1 P7 P3 CP5 T7 C3 FC5 F7 F3 AF3 Fp1 CENTRAL REGION LEFT HEMISPHERE RIGHT HEMISPHERE CENTRAL REGION LEFT HEMISPHERE RIGHT HEMISPHERE CENTRAL REGION LEFT HEMISPHERE RIGHT HEMISPHERE Fig. 2.…”

“…These potentials are generated as a result of a mismatch in the user's expected outcome of a task event and the actual one [2]. Since the first work suggesting that they could be detected in single trials [5], they have been successfully used within several BMI applications, such as for the correction of BMI outputs [6], classifier adaptation [7] or for learning sequential tasks [8], [9].…”

The Emotiv EPOC interface paradigm in Human-Computer Interaction

“…Ferrez and Millán [9], in 2008, Chavarriaga and Millán [19], in 2010, confirm that the ErrP signals are stable over time, that their shape is somewhat maintained for repeated measurements after an indefinite period of time -one year, for example -and then their triggering does not depend on the type of task but rather on the perception of incorrectness.…”

“…A second, more pronounced positive peak occurs at around 330 ms after the stimulus [19]. Although they come from the same sources as an EEG, the evoked potentials are much smaller than these (up to about 5 μV, while EEG waves can reach 50 μV).…”