The Adaptation Model Offers a Challenge for the Predictive Coding Account of Mismatch Negativity

Abstract

An unpredictable stimulus elicits a stronger event-related response than a high-probability stimulus. This differential in response magnitude is termed the mismatch negativity (MMN). Over the past decade, it has become increasingly popular to explain the MMN terms of predictive coding, a proposed general principle for the way the brain realizes Bayesian inference when it interprets sensory information. This perspective article is a reminder that the issue of MMN generation is far from settled, and that an alternative model in terms of adaptation continues to lurk in the wings. The adaptation model has been discounted because of the unrealistic and simplistic fashion in which it tends to be set up. Here, simulations of auditory cortex incorporating a modern version of the adaptation model are presented. These show that locally operating short-term synaptic depression accounts both for adaptation due to stimulus repetition and for MMN responses. This happens even in cases where adaptation has been ruled out as an explanation of the MMN (e.g., in the stimulus omission paradigm and the multi-standard control paradigm). Simulation models that would demonstrate the viability of predictive coding in a similarly multifaceted way are currently missing from the literature, and the reason for this is discussed in light of the current results.

Recurrent neural network model of human event-related potentials in response to intensity oddball stimulation

“…showing partial clustering of unit categories. This aligns with some of the arguments put forth by May (2021), namely that MMN is mistakenly interpreted as a separate component, and actually arises due to amplitude and latency modulation of neural generators that ordinarily precede the latency of MMN measurement. These findings are also consistent with a previous analysis of this dataset, which found that rising and falling sound level transitions produced opposite polarity changes in ERP amplitudes over the measurement window of MMN (O'Reilly, 2021b).…”

“…Changes in this signal are thought to reflect perceptual processing that can become impaired by underlying neuropathology, providing significant impetus for developing MMN as a clinical biomarker (Koshiyama et al, 2020; Schuelert et al, 2018; Taylor et al, 2017; Wang et al, 2022). However, considerable impediments to these efforts stem from ambiguity concerning the underlying neurophysiology of hypothetical prediction and prediction-error signalling components, and their dissociation from overlapping obligatory components that are sensitive to physical properties of sound (Heilbron and Chait, 2018; May, 2021; May and Tiitinen, 2010; O’Reilly and Conway, 2021; O’Reilly and O’Reilly, 2021). It has been suggested that MMN elicited by different physical changes in oddball stimuli may represent dissociable underlying neurophysiological processes (Näätänen et al, 2004; Nakajima et al, 2021; Pakarinen et al, 2007; Takegata et al, 2008; Todd et al, 2008), which perhaps implies a role for modulation of obligatory sensory components of the ERP by the physical properties of sounds in generating MMN (An et al, 2021; Hagenmuller et al, 2016; Kopp-Scheinpflug et al, 2018; O’Reilly, 2021a, 2021b; O’Reilly and Conway, 2021).…”

The Predictive Role of Low Spatial Frequencies in Automatic Face Processing: A Visual Mismatch Negativity Investigation

“…Accordingly, actual MMN paradigm, at least in the auditory domain, would not properly enable inferences about deviance detection and predictive coding because of state changes that can affect sensory responses. Even if our results are in line with previous findings and fit well with the predictive coding framework of MMN as well as with the predictive brain hypothesis, we cannot totally rule out that differences in sensory processing (see "Sensory Processing Theory" in O' Reilly and O'Reilly, 2021) or that another MMN framework such as the adaptation model (e.g., May, 2021) might partly explain our results. Indeed, despite intensively studied, MMN still poses a number of challenges in terms of interpretation (May, 2021).…”

“…The visual MMN (vMMN) is a differential negative event-related potential (ERP) representing the pre-attentive neural mechanism involved in the automatic detection of unpredicted visual changes among a learned regularity (Czigler et al, 2002 , 2006 ; Stefanics et al, 2015 ). In line with predictive coding framework (Rao and Ballard, 1999 ; Friston, 2005 , 2010 ), vMMN is often considered as a neural correlate of prediction error (Friston, 2005 ; Garrido et al, 2009 ; Stefanics et al, 2014 , but May, 2021 ; O'Reilly and O'Reilly, 2021 ) i.e., the difference between sensory input and predictions, based on our internal model constructed upon the regularity of standard stimuli. Thus, it appears to be particularly suitable to investigate predictive processes.…”

Decoding violated sensory expectations from the auditory cortex of anaesthetized mice: Hierarchical recurrent neural network depicts separate ‘danger’ and ‘safety’ units

“…Auditory novelty responses have been studied extensively in basic, clinical, and preclinical animal investigations, although despite this there remains substantial debate concerning their underlying neurophysiology (May, 2021;May and Tiitinen, 2010;Näätänen et al, 2005;.…”