Neural Correlates of Mnemonic and Predictive Representations in the Auditory Cortex


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date28 Sep 2022


In order for a sensory system to display adaptive behavior in response to external stimuli, it is advantageous to employ mechanisms capable of learning and encoding the probabilities present in ongoing stimulus streams. Because real-world stimuli tend to contain repetitive features, sensory systems have evolved to become highly sensitive to such regularities in order to detect deviations, while simultaneously "tuning out" stimuli features which do not require constant attendance owing to those regularities. This manner of efficient sequence processing relies on the ability to encode stimuli into short term memory, form predictions about upcoming stimulus features based on those that come before, and to make comparisons in order to update those assumptions when a deviant event is detected.

In this thesis, I investigate neural correlates during auditory sequence processing, used as a platform to probe and decode auditory sensory memories, predictions, and implicit learning in the auditory system. In the first experiment, I show that auditory sensory memory contents can be decoded from electrophysiological signals recorded in awake humans and anesthetized rats using homologous methods, suggesting that the mechanisms of sensory memory encoding are evolutionarily conserved across species. In the second experiment, I show that mnemonic and predictive representations of auditory stimuli can be simultaneously decoded from neural activity in anesthetized rats at overlapping latencies, but based on largely uncorrelated data features. Predictive representations are dynamically updated over the course of stimulation, suggesting a gradual formation of prediction. In the third experiment, conducted in awake humans, I show that neural correlates of prediction errors to unexpected sound contents are modulated by time-based predictions in a contextually congruent manner, such that local (vs. global) time-based predictions amplify prediction errors to unexpected sequence elements (vs. chunks). These modulations were shared between contextual levels in terms of the spatiotemporal distribution of neural activity, suggesting the brain integrates different predictions with a high degree of contextual specificity, but in a shared and distributed cortical network.

The experiments comprising this thesis explore phenomena that are integral to our understanding of cognitive processing and the mechanisms by which we interface with the outside world. As external stimuli contain incessant streams of complex regularities, the brain must find ways to parse meaningful information in the most efficient manner. The mechanisms responsible for this process rely on the brain’s intrinsic ability to learn such regularities, form a model allowing it to predict what events are likely to occur, and encode features into memory for comparison in order to update that model when deviants are detected. The following chapters detail the background of these mechanisms and three experiments which probe the resultant phenomena.

    Research areas

  • Auditory cortex