Decoding and Prediction of Sound Sequences in the Neocortex

皮層對聲音序列信息的解碼和預測

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date11 Feb 2022

Abstract

Sequence coding is essential for sensory processing. Encoding information about the probability of a given stimulus being presented, and about the probability of stimuli forming repeated segments, are two critical aspects to abstract the information from the sequences. It is thought that the mammalian auditory system is highly sensitive to the regularities of the sound sequence and able to use predictive processing in order to automatically detect unexpected (deviant) stimuli and segment sound streams. This thesis investigates the predictive processing of sound sequences in the cortex of rodents and humans based on complex sequence regularities.

The first study tests whether neural activity in the rat auditory cortex is modulated by previous segment experience. We recorded subdural responses using electrocorticography from the auditory cortex of 11 anesthetized rats. Prior to recording, four rats were trained to detect familiar triplets of acoustic stimuli (artificial syllables), three were passively exposed to the triplets, and another four rats had no training experience. While low-frequency neural activity was found to entrain (synchronize) to single stimuli, we did not find evidence for entrainment of neural activity to segments (triplets). However, in trained rats (but not in passively exposed and naïve rats), familiar triplets could be decoded more accurately than unfamiliar triplets based on distributed neural activity in the auditory cortex. These results suggest that rats process acoustic sequences and that the training experience modulates their cortical activity even under subsequent anesthesia.

In the second study, we tested whether the auditory cortical activity in awake mice shows sensitivity to violations of sequences based on local stimulus probability only or whether it is also sensitive to more complex violations based on stimulus order. We employed an auditory oddball paradigm, while recording wide-field calcium imaging from the auditory cortex of awake mice (N=8). During recording, mice were exposed to a series of standard pairs of artificial vowels and several types of deviant pairs. Prior to the oddball condition, we recorded the frequency response areas by presenting pure syllables at different sound levels, which enabled us to dissociate different auditory fields (A1, AAF, and A2). We found that mice could encode both the local probabilities and the more global stimulus patterns and elicited mismatch signals to the substitution deviants (pairs containing novel elements) and transposition deviants (pairs containing familiar elements in the wrong order), but not to omission deviants (pairs with elements missing). Notably, the A2 area elicited larger MMRs to those deviants than A1, which suggests a hierarchical gradient of prediction error signaling in the auditory system.

In the third study, we disambiguate neural correlates of “what” and “when” predictions by independently manipulating the predictability of temporal onset and acoustic contents at two hierarchical levels (single stimuli and stimulus pairs). Healthy volunteers (N=20) performed a repetition detection task while we recorded their neural activity using electroencephalogram. The results reveal that “what” and “when” predictions interactively modulated stimulus-evoked response amplitude in a hierarchically congruent manner, such that faster “when” predictions modulated the amplitude of mismatch responses to unexpected single stimuli, while slower “when” predictions modulated the amplitude of mismatch responses to unexpected stimulus pairs. We also find that the neural effects of these modulations were shared between the two hierarchical levels of prediction signalling in terms of the spatiotemporal distribution of EEG signals. Furthermore, by analyzing the entrainment of low-frequency neural activity to stimulus stream, we found evidence for a gradual increase of entrainment to slow temporal predictions (regarding the timing of stimulus pairs).

In conclusion, the thesis shows that rodents can encode both local probability and more complex and global patterns, such as perceiving as a chunk if they are highly trained. Besides, both rodents and humans can encode transitions and timing knowledge. Our results suggested that these potential mechanisms for sequence coding (i.e., transitions and timing, chunking) might be evolutionarily conserved in animals and humans.

    Research areas

  • ECoG, Predictive Coding, Sequence Learning, Rodents, Human EEG, Calcium Imaging