Brain signatures of surprise in EEG and MEG data

Abstract

10The brain is constantly anticipating the future of sensory inputs based on past 11 experiences. When new sensory data is different from predictions shaped by recent 12 trends, neural signals are generated to report this surprise. Existing models for 13 quantifying surprise are based on an ideal observer assumption operating under one 14 of the three definitions of surprise set forth as the Shannon, Bayesian, and 15 Confidence-corrected surprise. In this paper, we analyze both visual and auditory 16 EEG and auditory MEG signals recorded during oddball tasks to examine which 17 temporal components in these signals are sufficient to decode the brain's surprise 18 based on each of these three definitions. We found that for both recording systems 19 the Shannon surprise is always significantly better decoded than the Bayesian 20 surprise regardless of the sensory modality and the selected temporal features used 21 for decoding. 22 squared of regression model, ERP temporal components, generative distribution, 24 decoding power, middle temporal components, late temporal components, mismatch 25 negativity (MMN), P300, oddball task 26 Author summary 27 A regression model is proposed for decoding the level of the brain's surprise in 28 response to sensory sequences using selected temporal components of recorded EEG 3 29 and MEG data. Three surprise quantification definitions (Shannon, Bayesian,and 30 Confidence-corrected surprise) are compared in offering decoding power. Four 31 different regimes for selecting temporal samples of EEG and MEG data are used to 32 evaluate which part of the recorded data may contain signatures that represent the 33 brain's surprise in terms of offering a high decoding power. We found that both the 34 middle and late components of the EEG response offer strong decoding power for 35 surprise while the early components are significantly weaker in decoding surprise. In 36 the MEG response, we found that the middle components have the highest decoding 37 power while the late components offer moderate decoding powers. When using a 38 single temporal sample for decoding surprise, samples of the middle segment possess 39 the highest decoding power. Shannon surprise is always better decoded than the 40 other definitions of surprise for all the four temporal feature selection regimes. 41 Similar superiority for Shannon surprise is observed for the EEG and MEG data 42 across the entire range of temporal sample regimes used in our analysis. 43 45 next sensory input. Past inputs are used by the brain to form prior knowledge in the 46 Bayesian brain model [2,3]. In fact, the results of brain functions such as perception, 47 eliminating ambiguity, attention, and decision making are dependent on the way the 48 current sensory input and the knowledge gained from previous experiences are 49 combined in the hierarchical inference model of the brain [1] (for review see [4]). 50An input different from what the brain has predicted will be surprising in that it 51 generates a form of response measurable by brain ...

Neural surprise in somatosensory Bayesian learning

“…This interpretation is most intuitively compatible with the hypothesis that the early surprise signals may control later belief updating signals. Although the uncertainty regarding the exact half-lives is in line with the large variability found in the literature, local over global integration is consistently reported [9,13,39,48,94,95]. Given a fixed inter-stimulus interval of 750ms, a horizon of 95 and 26 observations may be equated to a half-life timescale of approximately 71 to 20 seconds, with regime switches expected to occur every 75 seconds.…”

“…Previous research either focused on their spatial identification using fMRI [11,45,46,47] or temporally specific, late EEG components [40]. Finally, to the best of our knowledge, only one recent pre-print study compared all three prominent surprise functions in a reanalysis of existing data, reporting PS to be better decoded across the entire post stimulus time-window [48].…”

Surprise: a unified theory and experimental predictions

“…Specifically: B. Standard generative model for studying passive learning in volatile environments (Adams & MacKay, 2007;Fearnhead & Liu, 2007;Liakoni et al, 2021;Nassar et al, 2012;Nassar et al, 2010;Wilson et al, 2013), C. Generative model corresponding to variants of bandit and reversal bandit tasks (Behrens et al, 2007;Findling et al, 2021;Horvath et al, 2021), where the cue variable X t = A t is a participant's action, D. Generative model for modeling human inferences about binary sequences (Gijsen et al, 2021;Maheu et al, 2019;Meyniel et al, 2016;Modirshanechi et al, 2019;Mousavi et al, 2020), and E. classic Markov Decision Processes (MDPs) (Daw et al, 2011;Gläscher et al, 2010;Huys et al, 2015;Lehmann et al, 2019;Schultz et al, 1997;Sutton & Barto, 2018), where the cue variable X t = (A t−1 , Y t−1 ) consists of previous action and observation. See Appendix A: Special cases and links to related works for details.…”

“…4 variable Z (e.g., the amplitude of the EEG P300 component (Kolossa et al, 2015;Meyniel et al, 2016)) is sensitive to or representative of surprise. Given two measures of surprise S and S , a typical experimental question is which one of them (if any) more accurately explains the variations of the variable Z (Gijsen et al, 2021;Kolossa et al, 2015;Mousavi et al, 2020;Ostwald et al, 2012;Visalli et al, 2021); see Fig. 2A1.…”

“…In the past decades, different definitions and formal measures of surprise have been proposed and studied (Baldi, 2002;Barto et al, 2013;Faraji et al, 2018;Friston, 2010;Gläscher et al, 2010;Itti & Baldi, 2006, 2009Kolossa et al, 2015;Liakoni et al, 2021;Palm, 2012;Schmidhuber, 2010;Shannon, 1948;Tribus, 1961). These surprise measures have been successful both in explaining the role of surprise in different brain functions (Findling et al, 2021;Friston, 2010;Gershman et al, 2017;Liakoni et al, 2021;Schmidhuber, 2010;Soltani & Izquierdo, 2019;Yu & Dayan, 2005) and in identifying signatures of surprise in behavioral and physiological measurements (Gijsen et al, 2021;Gläscher et al, 2010;Kolossa et al, 2015;Lieder et al, 2013;Maheu et al, 2019;Mars et al, 2008;Meyniel, 2020;Meyniel et al, 2016;Modirshanechi et al, 2019;Mousavi et al, 2020;Ostwald et al, 2012;Preuschoff et al, 2011;Rubin et al, 2016). However, there are still many open questions (Baldi, 2002;Barto et al, 2013;Liakoni et al, 2021;Palm, 2012;Schmidhuber, 2010) including, but not limited to: (i) Since the notion of surprise has been linked to a multitude of computational roles and physiological phenomena, should we st...…”