Predictive coding posits that sensory systems operate under hierarchical Bayesian principles and perception is achieved by minimizing prediction errors. The present dissertation aimed to investigate the neural correlates of predictive coding quantities through a multifaceted approach, which involved meta-analytically summarizing the available literature on predictive coding and utilizing model-based electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) to provide new insights regarding the foundational mechanisms hypothesized by this framework. Chapter 2’s results support the hypothesis that predictive coding units operate within a domain-general, large-scale network. Consistent activations were observed across diverse cognitive domains and prediction conditions, underscoring the contribution of several key hubs. This widespread representation can be attributed to the continuous generation of predictions and prediction errors by the brain, whether engaged in an experimental task or interacting with the external environment or internal world. If framed within a heterarchical structure, such an organization would allow for prediction units operating within different sets of brain areas to be recruited depending on the brain state prior to receiving an input, or on the context, in accordance with predictive coding-consistent degeneracy models. Chapter 3’s results supported the hypothesis that trial-by-trial EEG activity would reflect the evolution of precision-weighted prediction error (pwPE) trajectories during early processing stages of complex, socially relevant stimuli, such as faces. However, it did not provide evidence for, nor could it refute, the hypothesis that pwPEs reflected in the trial-by-trial neural signal are differentially modulated by the valence of expressions deviating from expectations. Overall, this study demonstrated that the predictive coding principles applicable to stimuli yielding specific predictions at the level of V1 receptive fields also broadly extend to the sensory system’s mechanisms for detecting and processing feature-selective deviations from expectations in complex stimuli, such as facial expressions. The study highlights how neural responses to facial expressions are contingent on the fit between ongoing predictions and actual sensory input. Chapter 4’s results supported the hypothesis that the anterior insula (AI) is involved in feature-level (i.e., view-independent familiarity) updating, error signaling, and context-level familiarity updating to reach recognition judgments, thereby guiding perceptual decision-making. A significant increase in BOLD percentage signal change was observed in both the left and right AI as a function of computationally derived predictive coding measures, suggesting that AI activity reflects its role in encoding the strength of an agent's belief, as well as the trial-by-trial updating and refinement of those beliefs in response to the statistical dependencies within their environment. The results also underscore the AI’s involvement in learning and recognition judgment processes. Collectively, these results provide insights into the complex interplay between predictions and sensory information in the brain, reiterating that perception is not merely a passive reflection of incoming stimuli but a dynamic, adaptive process influenced by our expectations. Lastly, Chapter 5 introduces an important instrument for the cognitive psychology and neuropsychology scientific community: the PECANS checklist, a tool designed to enhance research rigor, improve the completeness of scientific reporting, and promote reproducibility.

NEURAL CORRELATES OF PREDICTIVE CODING QUANTITIES

COSTA, CRISTIANO
2025

Abstract

Predictive coding posits that sensory systems operate under hierarchical Bayesian principles and perception is achieved by minimizing prediction errors. The present dissertation aimed to investigate the neural correlates of predictive coding quantities through a multifaceted approach, which involved meta-analytically summarizing the available literature on predictive coding and utilizing model-based electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) to provide new insights regarding the foundational mechanisms hypothesized by this framework. Chapter 2’s results support the hypothesis that predictive coding units operate within a domain-general, large-scale network. Consistent activations were observed across diverse cognitive domains and prediction conditions, underscoring the contribution of several key hubs. This widespread representation can be attributed to the continuous generation of predictions and prediction errors by the brain, whether engaged in an experimental task or interacting with the external environment or internal world. If framed within a heterarchical structure, such an organization would allow for prediction units operating within different sets of brain areas to be recruited depending on the brain state prior to receiving an input, or on the context, in accordance with predictive coding-consistent degeneracy models. Chapter 3’s results supported the hypothesis that trial-by-trial EEG activity would reflect the evolution of precision-weighted prediction error (pwPE) trajectories during early processing stages of complex, socially relevant stimuli, such as faces. However, it did not provide evidence for, nor could it refute, the hypothesis that pwPEs reflected in the trial-by-trial neural signal are differentially modulated by the valence of expressions deviating from expectations. Overall, this study demonstrated that the predictive coding principles applicable to stimuli yielding specific predictions at the level of V1 receptive fields also broadly extend to the sensory system’s mechanisms for detecting and processing feature-selective deviations from expectations in complex stimuli, such as facial expressions. The study highlights how neural responses to facial expressions are contingent on the fit between ongoing predictions and actual sensory input. Chapter 4’s results supported the hypothesis that the anterior insula (AI) is involved in feature-level (i.e., view-independent familiarity) updating, error signaling, and context-level familiarity updating to reach recognition judgments, thereby guiding perceptual decision-making. A significant increase in BOLD percentage signal change was observed in both the left and right AI as a function of computationally derived predictive coding measures, suggesting that AI activity reflects its role in encoding the strength of an agent's belief, as well as the trial-by-trial updating and refinement of those beliefs in response to the statistical dependencies within their environment. The results also underscore the AI’s involvement in learning and recognition judgment processes. Collectively, these results provide insights into the complex interplay between predictions and sensory information in the brain, reiterating that perception is not merely a passive reflection of incoming stimuli but a dynamic, adaptive process influenced by our expectations. Lastly, Chapter 5 introduces an important instrument for the cognitive psychology and neuropsychology scientific community: the PECANS checklist, a tool designed to enhance research rigor, improve the completeness of scientific reporting, and promote reproducibility.
12-feb-2025
Inglese
SCARPAZZA, CRISTINA
Università degli studi di Padova
File in questo prodotto:
File Dimensione Formato  
PhD Thesis_Definitive_Costa Cristiano.pdf

accesso aperto

Dimensione 17.8 MB
Formato Adobe PDF
17.8 MB Adobe PDF Visualizza/Apri

I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.14242/194946
Il codice NBN di questa tesi è URN:NBN:IT:UNIPD-194946