Predictive Coding Model Research Articles

Disturbances of Interoception in Schizophrenia – Focus on Neuroimaging

Abstract: The current paper presents the most recent theoretical understanding of interoception (INTC) in light of the predictive processing framework and the Embodied Predictive Interoceptive Coding model and reviews recent experimental data on different interoceptive disturbances in schizophrenia (SCZ) such as lower interoceptive accuracy, nociceptive dysfunction reflected by reduced pain sensitivity, disturbed thermoregulation, impaired gastrointestinal regulation, cardiovascular and respiratory dysfunction, and immune disturbances. The immense clinical implications of these disturbances are discussed in the context of the high rates of morbidity and mortality in SCZ related to suicide and cardiac death. Next, the potential of INTC-related treatments is reviewed, and conclusions are made on the evidence for the use of vagus nerve stimulation, mindfulness-based interventions, and exercise therapy. The neuroimaging perspective is given in a brief account of the INTC-related findings in healthy individuals suggesting the right anterior insula and anterior cingulate cortex as key brain areas for interoceptive awareness, accuracy, and sensibility, by supporting higher-order representation and integration of visceral, and somatosensory interoception with exteroceptive information. In SCZ, on the other hand, neuroimaging research on specific interoceptive abilities is lacking but indirect evidence points to disrupted activation of major INTC network hubs. There is converging evidence of changes in these hubs in terms of both structure and function (volume reductions, aberrant activation, and connectivity). This narrative review identifies several gaps in contemporary INTC-related research in SCZ that need to be addressed in future experimental and clinical studies to elucidate the exact nature of the disturbances of interoceptive abilities in modalities other than cardioception; the interactions between different interoceptive modalities; the association between interoceptive abilities and other mental functions (such as emotion recognition and regulation, decision making, learning, memory, etc.); the efficacy of nonpharmacological and pharmacological INTC-based interventions; and the relation of disturbed interoceptive abilities with brain structure and function.

Learning to segment self-generated from externally caused optic flow through sensorimotor mismatch circuits

Efficient sensory detection requires the capacity to ignore task-irrelevant information, for example when optic flow patterns created by egomotion need to be disentangled from object perception. To investigate how this is achieved in the visual system, predictive coding with sensorimotor mismatch detection is an attractive starting point. Indeed, experimental evidence for sensorimotor mismatch signals in early visual areas exists, but it is not understood how they are integrated into cortical networks that perform input segmentation and categorization. Our model advances a biologically plausible solution by extending predictive coding models with the ability to distinguish self-generated from externally caused optic flow. We first show that a simple three neuron circuit produces experience-dependent sensorimotor mismatch responses, in agreement with calcium imaging data from mice. This microcircuit is then integrated into a neural network with two generative streams. The motor-to-visual stream consists of parallel microcircuits between motor and visual areas and learns to spatially predict optic flow resulting from self-motion. The second stream bidirectionally connects a motion-selective higher visual area (mHVA) to V1, assigning a crucial role to the abundant feedback connections to V1: the maintenance of a generative model of externally caused optic flow. In the model, area mHVA learns to segment moving objects from the background, and facilitates object categorization. Based on shared neurocomputational principles across species, the model also maps onto primate visual cortex. Our work extends Hebbian predictive coding to sensorimotor settings, in which the agent actively moves - and learns to predict the consequences of its own movements.

How Sports Involvement and Brand Fit Influence the Effectiveness of Sports Sponsorship from the Perspective of Predictive Coding Theory: An Event-Related Potential (ERP)-Based Study.

Background/Objectives: With the rapid expansion of the global sports market, the significance of sports sponsorship has attracted growing attention. However, during the golden age of the sports industry's development in China, international sports brand giants such as Nike, Adidas, and Under Armour have rapidly captured a substantial share of the Chinese sports consumer market through their distinctive product designs and varied marketing strategies. This has resulted in a highly competitive environment for China's sports goods industry. Therefore, fostering the improved development of domestic sports brands has become a crucial issue deserving of thorough scholarly investigation. This study examines how consumers' differing levels of sports involvement and the degree of fit between the sponsoring brand and the sponsored event affect their cognitive and emotional responses to sports sponsorships. Methods: By employing Predictive Coding Theory and ERP (event-related potential) brainwave technology, this study delves into the psychological and neurobiological levels to analyze the impact of consumer sports involvement on the processing of sponsorship information. Results: The results indicate significant differences in cognitive and emotional responses between high-involvement and low-involvement consumers. Additionally, the fit between the sponsoring brand and the sponsored event also significantly affects consumers' cognitive and emotional responses. These differences stem from consumers' complex and sophisticated predictive coding models. Conclusions: This study not only provides scientific evidence for sports brands in selecting and executing sponsorship activities, but also offers new perspectives for evaluating and optimizing sponsorship effectiveness.

Collective predictive coding hypothesis: symbol emergence as decentralized Bayesian inference.

Understanding the emergence of symbol systems, especially language, requires the construction of a computational model that reproduces both the developmental learning process in everyday life and the evolutionary dynamics of symbol emergence throughout history. This study introduces the collective predictive coding (CPC) hypothesis, which emphasizes and models the interdependence between forming internal representations through physical interactions with the environment and sharing and utilizing meanings through social semiotic interactions within a symbol emergence system. The total system dynamics is theorized from the perspective of predictive coding. The hypothesis draws inspiration from computational studies grounded in probabilistic generative models and language games, including the Metropolis-Hastings naming game. Thus, playing such games among agents in a distributed manner can be interpreted as a decentralized Bayesian inference of representations shared by a multi-agent system. Moreover, this study explores the potential link between the CPC hypothesis and the free-energy principle, positing that symbol emergence adheres to the society-wide free-energy principle. Furthermore, this paper provides a new explanation for why large language models appear to possess knowledge about the world based on experience, even though they have neither sensory organs nor bodies. This paper reviews past approaches to symbol emergence systems, offers a comprehensive survey of related prior studies, and presents a discussion on CPC-based generalizations. Future challenges and potential cross-disciplinary research avenues are highlighted.

Energy-information trade-off induces continuous and discontinuous phase transitions in lateral predictive coding

Lateral predictive coding is a recurrent neural network that creates energy-efficient internal representations by exploiting statistical regularity in sensory inputs. Here, we analytically investigate the trade-off between information robustness and energy in a linear model of lateral predictive coding and numerically minimize a free energy quantity. We observed several phase transitions in the synaptic weight matrix, particularly a continuous transition that breaks reciprocity and permutation symmetry and builds cyclic dominance and a discontinuous transition with the associated sudden emergence of tight balance between excitatory and inhibitory interactions. The optimal network follows an ideal gas law over an extended temperature range and saturates the efficiency upper bound of energy use. These results provide theoretical insights into the emergence and evolution of complex internal models in predictive processing systems.

Dissecting Mismatch Negativity: Early and Late Subcomponents for Detecting Deviants in Local and Global Sequence Regularities.

Mismatch negativity (MMN) is commonly recognized as a neural signal of prediction error evoked by deviants from the expected patterns of sensory input. Studies show that MMN diminishes when sequence patterns become more predictable over a longer timescale. This implies that MMN is composed of multiple subcomponents, each responding to different levels of temporal regularities. To probe the hypothesized subcomponents in MMN, we record human electroencephalography during an auditory local-global oddball paradigm where the tone-to-tone transition probability (local regularity) and the overall sequence probability (global regularity) are manipulated to control temporal predictabilities at two hierarchical levels. We find that the size of MMN is correlated with both probabilities and the spatiotemporal structure of MMN can be decomposed into two distinct subcomponents. Both subcomponents appear as negative waveforms, with one peaking early in the central-frontal area and the other late in a more frontal area. With a quantitative predictive coding model, we map the early and late subcomponents to the prediction errors that are tied to local and global regularities, respectively. Our study highlights the hierarchical complexity of MMN and offers an experimental and analytical platform for developing a multitiered neural marker applicable in clinical settings.

Predictive coding networks for temporal prediction.

One of the key problems the brain faces is inferring the state of the world from a sequence of dynamically changing stimuli, and it is not yet clear how the sensory system achieves this task. A well-established computational framework for describing perceptual processes in the brain is provided by the theory of predictive coding. Although the original proposals of predictive coding have discussed temporal prediction, later work developing this theory mostly focused on static stimuli, and key questions on neural implementation and computational properties of temporal predictive coding networks remain open. Here, we address these questions and present a formulation of the temporal predictive coding model that can be naturally implemented in recurrent networks, in which activity dynamics rely only on local inputs to the neurons, and learning only utilises local Hebbian plasticity. Additionally, we show that temporal predictive coding networks can approximate the performance of the Kalman filter in predicting behaviour of linear systems, and behave as a variant of a Kalman filter which does not track its own subjective posterior variance. Importantly, temporal predictive coding networks can achieve similar accuracy as the Kalman filter without performing complex mathematical operations, but just employing simple computations that can be implemented by biological networks. Moreover, when trained with natural dynamic inputs, we found that temporal predictive coding can produce Gabor-like, motion-sensitive receptive fields resembling those observed in real neurons in visual areas. In addition, we demonstrate how the model can be effectively generalized to nonlinear systems. Overall, models presented in this paper show how biologically plausible circuits can predict future stimuli and may guide research on understanding specific neural circuits in brain areas involved in temporal prediction.

Understanding and Improving Optimization in Predictive Coding Networks

Backpropagation (BP), the standard learning algorithm for artificial neural networks, is often considered biologically implausible. In contrast, the standard learning algorithm for predictive coding (PC) models in neuroscience, known as the inference learning algorithm (IL), is a promising, bio-plausible alternative. However, several challenges and questions hinder IL's application to real-world problems. For example, IL is computationally demanding, and without memory-intensive optimizers like Adam, IL may converge to poor local minima. Moreover, although IL can reduce loss more quickly than BP, the reasons for these speedups or their robustness remains unclear. In this paper, we tackle these challenges by 1) altering the standard implementation of PC circuits to substantially reduce computation, 2) developing a novel optimizer that improves the convergence of IL without increasing memory usage, and 3) establishing theoretical results that help elucidate the conditions under which IL is sensitive to second and higher-order information.

Temporal imprecision of phase coherence in schizophrenia and psychosis-dynamic mechanisms and diagnostic marker.

Schizophrenia (SCZ) is a complex disorder in which various pathophysiological models have been postulated. Brain imaging studies using EEG/MEG and fMRI show altered amplitude and, more recently, decrease in phase coherence in response to external stimuli. What are the dynamic mechanisms of such phase incoherence, and can it serve as a differential-diagnostic marker? Addressing this gap in our knowledge, we uniquely combine a review of previous findings, novel empirical data, and computational-dynamic simulation. The main findings are: (i) the review shows decreased phase coherence in SCZ across a variety of different tasks and frequencies, e.g., task- and frequency-unspecific, which is further supported by our own novel data; (ii) our own data demonstrate diagnostic specificity of decreased phase coherence for SCZ as distinguished from major depressive disorder; (iii) simulation data exhibit increased phase offset in SCZ leading to a precision index, in the millisecond range, of the phase coherence relative to the timing of the external stimulus. Together, we demonstrate the key role of temporal imprecision in phase coherence of SCZ, including its mechanisms (phase offsets, precision index) on the basis of which we propose a phase-based temporal imprecision model of psychosis (PTP). The PTP targets a deeper dynamic layer of a basic disturbance. This converges well with other models of psychosis like the basic self-disturbance and time-space experience changes, as discussed in phenomenological and spatiotemporal psychopathology, as well as with the models of aberrant predictive coding and disconnection as in computational psychiatry. Finally, our results show that temporal imprecision as manifest in decreased phase coherence is a promising candidate biomarker for clinical differential diagnosis of SCZ, and more broadly, psychosis.

Abnormalities of structural covariance of insular subregions in drug-naïve OCD patients.

The insula plays a significant role in the neural mechanisms of obsessive-compulsive disorder. Previous studies have identified functional and structural abnormalities in insula in obsessive-compulsive disorder patients. The predictive coding model in the context of interoception can explain the psychological and neuropathological manifestations observed in obsessive-compulsive disorder. The model is based on the degree of laminar differentiation of cerebral cortex. The interindividual differences in a local measure of brain structure often covary with interindividual differences in other brain regions. We investigated the anatomical network involving the insula in a drug-naïve obsessive-compulsive disorder sample. We recruited 58 obsessive-compulsive disorder patients and 84 matched health controls. The cortical thickness covariance maps between groups were compared at each vertex. We also evaluated the modulation of Yale-Brown Obsessive-Compulsive Scale scores and obsessive-compulsive disorder duration on thickness covariance. Our findings indicated that the thickness covariance seeded from granular and dysgranular insula are different compared with controls. The duration and severity of obsessive-compulsive disorder can modulate the thickness covariance of granular and dysgranular insula with posterior cingulate cortex and rostral anterior cingulate cortex. Our results revealed aberrant insular structural characteristics and cortical thickness covariance in obsessive-compulsive disorder patients, contributing to a better understanding of the involvement of insula in the pathological mechanisms underlying obsessive-compulsive disorder.

Active Predictive Coding: A Unifying Neural Model for Active Perception, Compositional Learning, and Hierarchical Planning.

There is growing interest in predictive coding as a model of how the brain learns through predictions and prediction errors. Predictive coding models have traditionally focused on sensory coding and perception. Here we introduce active predictive coding (APC) as a unifying model for perception, action, and cognition. The APC model addresses important open problems in cognitive science and AI, including (1) how we learn compositional representations (e.g., part-whole hierarchies for equivariant vision) and (2) how we solve large-scale planning problems, which are hard for traditional reinforcement learning, by composing complex state dynamics and abstract actions from simpler dynamics and primitive actions. By using hypernetworks, self-supervised learning, and reinforcement learning, APC learns hierarchical world models by combining task-invariant state transition networks and task-dependent policy networks at multiple abstraction levels. We illustrate the applicability of the APC model to active visual perception and hierarchical planning. Our results represent, to our knowledge, the first proof-of-concept demonstration of a unified approach to addressing the part-whole learning problem in vision, the nested reference frames learning problem in cognition, and the integrated state-action hierarchy learning problem in reinforcement learning.

Perspective on Emotion from Allostasis Mediated by Interoception

Feeling emotional responses, such as heartbeat and respiration, influence subjective feeling of emotion. The role of interoception as a link in this relationship has been the focus of much attention. The predictive coding model is suggested for understanding the mechanism of interoception, and the model is also closely related to allostasis. This paper introduces a perspective on emotion from the allostasis mediated by interoception.

The effect of multisensory semantic congruency on unisensory object recognition in schizophrenia.

Multisensory, as opposed to unisensory processing of stimuli, has been found to enhance the performance (e.g., reaction time, accuracy, and discrimination) of healthy individuals across various tasks. However, this enhancement is not as pronounced in patients with schizophrenia (SZ), indicating impaired multisensory integration (MSI) in these individuals. To the best of our knowledge, no study has yet investigated the impact of MSI deficits in the context of working memory, a domain highly reliant on multisensory processing and substantially impaired in schizophrenia. To address this research gap, we employed two adopted versions of the continuous object recognition task to investigate the effect of single-trail multisensory encoding on subsequent object recognition in 21 schizophrenia patients and 21 healthy controls (HC). Participants were tasked with discriminating between initial and repeated presentations. For the initial presentations, half of the stimuli were audiovisual pairings, while the other half were presented unimodal. The task-relevant stimuli were then presented a second time in a unisensory manner (either auditory stimuli in the auditory task or visual stimuli in the visual task). To explore the impact of semantic context on multisensory encoding, half of the audiovisual pairings were selected to be semantically congruent, while the remaining pairs were not semantically related to each other. Consistent with prior studies, our findings demonstrated that the impact of single-trial multisensory presentation during encoding remains discernible during subsequent object recognition. This influence could be distinguished based on the semantic congruity between the auditory and visual stimuli presented during the encoding. This effect was more robust in the auditory task. In the auditory task, when congruent multisensory pairings were encoded, both participant groups demonstrated a multisensory facilitation effect. This effect resulted in improved accuracy and RT performance. Regarding incongruent audiovisual encoding, as expected, HC did not demonstrate an evident multisensory facilitation effect on memory performance. In contrast, SZs exhibited an atypically accelerated reaction time during the subsequent auditory object recognition. Based on the predictive coding model we propose that this observed deviations indicate a reduced semantic modulatory effect and anomalous predictive errors signaling, particularly in the context of conflicting cross-modal sensory inputs in SZ.

An Introduction to Predictive Processing Models of Perception and Decision-Making.

The predictive processing framework includes a broad set of ideas, which might be articulated and developed in a variety of ways, concerning how the brain may leverage predictive models when implementing perception, cognition, decision-making, and motor control. This article provides an up-to-date introduction to the two most influential theories within this framework: predictive coding and active inference. The first half of the paper (Sections 2-5) reviews the evolution of predictive coding, from early ideas about efficient coding in the visual system to a more general model encompassing perception, cognition, and motor control. The theory is characterized in terms of the claims it makes at Marr's computational, algorithmic, and implementation levels of description, and the conceptual and mathematical connections between predictive coding, Bayesian inference, and variational free energy (a quantity jointly evaluating model accuracy and complexity) are explored. The second half of the paper (Sections 6-8) turns to recent theories of active inference. Like predictive coding, active inference models assume that perceptual and learning processes minimize variational free energy as a means of approximating Bayesian inference in a biologically plausible manner. However, these models focus primarily on planning and decision-making processes that predictive coding models were not developed to address. Under active inference, an agent evaluates potential plans (action sequences) based on their expected free energy (a quantity that combines anticipated reward and information gain). The agent is assumed to represent the world as a partially observable Markov decision process with discrete time and discrete states. Current research applications of active inference models are described, including a range of simulation work, as well as studies fitting models to empirical data. The paper concludes by considering future research directions that will be important for further development of both models.

Dynamic Bayesian Contrastive Predictive Coding Model for Personalized Product Search

In this article, we study the problem of dynamic personalized product search. Due to the data-sparsity problem in the real world, existing methods suffer from the challenge of data inefficiency. We address the challenge by proposing a Dynamic Bayesian Contrastive Predictive Coding model (DBCPC), which aims to capture the rich structured information behind search records to improve data efficiency. Our proposed DBCPC utilizes contrastive predictive learning to jointly learn dynamic embeddings with structure information of entities (i.e., users, products, and words). Specifically, our DBCPC employs structured prediction to tackle the intractability caused by non-linear output space and utilizes the time embedding technique to avoid designing different encoders each time in the Dynamic Bayesian models. In this way, our model jointly learns the underlying embeddings of entities (i.e., users, products, and words) via prediction tasks, which enables the embeddings to focus more on their general attributes and capture the general information during the preference evolution with time. For inferring the dynamic embeddings, we propose an inference algorithm combining the variational objective and the contrastive objectives. Experiments were conducted on an Amazon dataset and the experimental results show that our proposed DBCPC can learn the higher-quality embeddings and outperforms the state-of-the-art non-dynamic and dynamic models for product search.

Online Intention Recognition With Incomplete Information Based on a Weighted Contrastive Predictive Coding Model in Wargame.

The incomplete and imperfect essence of the battlefield situation results in a challenge to the efficiency, stability, and reliability of traditional intention recognition methods. For this problem, we propose a deep learning architecture that consists of a contrastive predictive coding (CPC) model, a variable-length long short-term memory network (LSTM) model, and an attention weight allocator for online intention recognition with incomplete information in wargame (W-CPCLSTM). First, based on the typical characteristics of intelligence data, a CPC model is designed to capture more global structures from limited battlefield information. Then, a variable-length LSTM model is employed to classify the learned representations into predefined intention categories. Next, a weighted approach to the training attention of CPC and LSTM is introduced to allow for the stability of the model. Finally, performance evaluation and application analysis of the proposed model for the online intention recognition task were carried out based on four different degrees of detection information and a perfect situation of ideal conditions in a wargame. Besides, we explored the effect of different lengths of intelligence data on recognition performance and gave application examples of the proposed model to a wargame platform. The simulation results demonstrate that our method not only contributes to the growth of recognition stability, but it also improves recognition accuracy by 7%-11%, 3%-7%, 3%-13%, and 3%-7%, the recognition speed by 6-32x, 4-18x, 13-*x, and 1-6x compared with the traditional LSTM, classical FCN, OctConv, and OctFCN models, respectively, which characterizes it as a promising reference tool for command decision-making.

Serial Dependence in Perception.

Much evidence has shown that perception is biased towards previously presented similar stimuli, an effect recently termed serial dependence. Serial dependence affects nearly every aspect of perception, often causing gross perceptual distortions, especially for weak and ambiguous stimuli. Despite unwanted side-effects, empirical evidence and Bayesian modeling show that serial dependence acts to improve efficiency and is generally beneficial to the system. Consistent with models of predictive coding, the Bayesian priors of serial dependence are generated at high levels of cortical analysis, incorporating much perceptual experience, but feed back to lower sensory areas. These feedback loops may drive oscillations in the alpha range, linked strongly with serial dependence. The discovery of top-down predictive perceptual processes is not new, but the new, more quantitative approach characterizing serial dependence promises to lead to a deeper understanding of predictive perceptual processes and their underlying neural mechanisms.

Invited Session III: Neural network models of the visual system: Robust information representation in hierarchical networks of the visual cortex.

The mammalian visual system has an exquisite ability to encode relevant information from noisy sensory inputs in natural scenes. Such robust yet adaptable computation is believed to emerge from the uniquely complex connectivity structure of biological neural networks. In this talk, I will focus on the hierarchical organization of the cortical network, and discuss how this unique network structure shapes robust representations of noisy visual inputs via predictive coding. Using a predictive coding model of hierarchically-related visual cortical areas, we link feedback connections to top-down predictions of the lower cortical neural activity. Our model shows that hierarchical predictive coding explains the mechanisms of robust recognition of noisy objects in intermediate cortical areas. Secondly, we study how context-specific information may be represented and modified across learning in populations of neurons in a recurrent neural network model of hierarchically-related cortical areas. Motivated by an experimental study investigating neural activities in response to expected and unexpected natural images in mouse visual cortex, we test how the divergence of feedforward and feedback connections underlie differential representations of expected and surprising sensory information across cortical areas and layers. In sum, our study provides insights into how robust and adaptable visual encoding arise from biologically-motivated connectivity across the cortical hierarchy.

Stimulus expectations do not modulate visual event-related potentials in probabilistic cueing designs

Humans and other animals can learn and exploit repeating patterns that occur within their environments. These learned patterns can be used to form expectations about future sensory events. Several influential predictive coding models have been proposed to explain how learned expectations influence the activity of stimulus-selective neurons in the visual system. These models specify reductions in neural response measures when expectations are fulfilled (termed expectation suppression) and increases following surprising sensory events. However, there is currently scant evidence for expectation suppression in the visual system when confounding factors are taken into account. Effects of surprise have been observed in blood oxygen level dependent (BOLD) signals, but not when using electrophysiological measures. To provide a strong test for expectation suppression and surprise effects we performed a predictive cueing experiment while recording electroencephalographic (EEG) data. Participants (n=48) learned cue-face associations during a training session and were then exposed to these cue-face pairs in a subsequent experiment. Using univariate analyses of face-evoked event-related potentials (ERPs) we did not observe any differences across expected (90% probability), neutral (50%) and surprising (10%) face conditions. Across these comparisons, Bayes factors consistently favoured the null hypothesis throughout the time-course of the stimulus-evoked response. When using multivariate pattern analysis we did not observe above-chance classification of expected and surprising face-evoked ERPs. By contrast, we found robust within- and across-trial stimulus repetition effects. Our findings do not support predictive coding-based accounts that specify reduced prediction error signalling when perceptual expectations are fulfilled. They instead highlight the utility of other types of predictive processing models that describe expectation-related phenomena in the visual system without recourse to prediction error signalling.

Hybrid predictive coding: Inferring, fast and slow.

Predictive coding is an influential model of cortical neural activity. It proposes that perceptual beliefs are furnished by sequentially minimising "prediction errors"-the differences between predicted and observed data. Implicit in this proposal is the idea that successful perception requires multiple cycles of neural activity. This is at odds with evidence that several aspects of visual perception-including complex forms of object recognition-arise from an initial "feedforward sweep" that occurs on fast timescales which preclude substantial recurrent activity. Here, we propose that the feedforward sweep can be understood as performing amortized inference (applying a learned function that maps directly from data to beliefs) and recurrent processing can be understood as performing iterative inference (sequentially updating neural activity in order to improve the accuracy of beliefs). We propose a hybrid predictive coding network that combines both iterative and amortized inference in a principled manner by describing both in terms of a dual optimization of a single objective function. We show that the resulting scheme can be implemented in a biologically plausible neural architecture that approximates Bayesian inference utilising local Hebbian update rules. We demonstrate that our hybrid predictive coding model combines the benefits of both amortized and iterative inference-obtaining rapid and computationally cheap perceptual inference for familiar data while maintaining the context-sensitivity, precision, and sample efficiency of iterative inference schemes. Moreover, we show how our model is inherently sensitive to its uncertainty and adaptively balances iterative and amortized inference to obtain accurate beliefs using minimum computational expense. Hybrid predictive coding offers a new perspective on the functional relevance of the feedforward and recurrent activity observed during visual perception and offers novel insights into distinct aspects of visual phenomenology.