Increased Attentive Use Is Linked to More Idiosyncratic Functional Connections Experience is thought to modify neural connections to adapt the network to be more optimal for the environment. Given the brain’s complexity, multiple network changes could each move the system toward optimality. Standard approaches examine each connection independently; these studies have often shown considerable interindividual variability and modest effects (Marek et al., 2022). Here, we take a different strategy, determining how a whole-brain connection pattern differs from the typical pattern, that is, how “idiosyncratic” the pattern is. We examined how the idiosyncrasy of whole-brain connection patterns varies with frequency of the use of that part of cortex for attention-demanding tasks, focusing on central versus peripheral vision in healthy individuals (who use central vision more frequently for attention-demanding tasks). We found that the whole-brain pattern of functional connections to the cortical representations of central vision is idiosyncratic, whereas patterns of connections to representations of peripheral vision were very similar person to person in healthy vision controls (14 females, 9 males). In a second set of analyses, we examined the brains of people with central vision loss (11 females, 10 males) who use a portion of peripheral vision [the preferred retinal locus (PRL)] more frequently for attention-demanding tasks in their daily lives. The cortical representation of the PRL exhibits more idiosyncratic connections, compared with a control brain region or compared with the same brain region in matched healthy vision controls. These results are consistent with the hypothesis that increased attentive use of a brain area results in idiosyncratic patterns of whole-brain connections.
Active inference and artificial reasoning This paper considers the sampling of outcomes that provide the greatest amount of information about the structure of underlying world models. This generalisation furnishes a principled approach to structure learning, and specifically rule discovery, under a plausible set of generative models or hypotheses. In active inference, policies—i.e., combinations of actions—are selected based on their expected free energy, which comprises expected information gain and value. Information gain corresponds to the Kullback-Leibler divergence between predictive posteriors with, and without, the consequences of action. Posteriors over models can be evaluated quickly and efficiently using Bayesian Model Reduction, based upon accumulated posterior beliefs about model parameters. The ensuing information gain can then be used to select actions that disambiguate among alternative models, in the spirit of optimal experimental design. We illustrate this kind of active selection or reasoning using partially observed discrete models; namely, a three-ball paradigm used previously to describe artificial insight and aha moments via (synthetic) introspection or sleep. We focus on the sample efficiency afforded by seeking outcomes that resolve the greatest uncertainty about the world model, under which outcomes are generated.
Hierarchical Active Inference Using Successor Representations Active inference, a neurally inspired model for inferring actions based on the free energy principle (FEP), has been proposed as a unifying framework for understanding perception, action, and learning in the brain. Active inference has previously been used to model ecologically important tasks such as navigation and planning, but scaling it to solve complex large-scale problems in real-world environments has remained a challenge. Inspired by the existence of multiscale hierarchical representations in the brain, we propose a model for planning actions based on hierarchical active inference. Our approach combines a hierarchical model of the environment with successor representations for efficient planning. We present results demonstrating (1) how lower-level successor representations can be used to learn higher-level abstract states, (2) how planning based on active inference at the lower level can be used to bootstrap and learn higher-level abstract actions, and (3) how these learned higher-level abstract states and actions can facilitate efficient planning. We illustrate the performance of the approach on several planning and reinforcement learning (RL) problems, including a variant of the well-known four rooms task, a key-based navigation task, a partially observable planning problem, the mountain car problem, and PointMaze, a family of navigation tasks with continuous state and action spaces. Our results represent, to our knowledge, the first application of learned hierarchical state and action abstractions to active inference in FEP-based theories of brain function.
Identifiability of Bayesian models of perception Inferring the underlying computational processes from behavioral measurements is a fundamental approach in cognitive science and neuroscience. Although Bayesian decision theory has become a major normative framework for modeling perception and cognition, it is unclear to what extent its modeling components (i.e., prior belief, likelihood function, and loss function) can be recovered from behavioral data. Here, we systematically investigated the problem of inferring such Bayesian models from behavioral tasks. We determined the situations under which some components of Bayesian models are systematically confounded, as well as the practical choices in experimental design that can resolve such ambiguity. Overall, our analytical results guarantee in-principle identifiability under broadly applicable conditions, without any a priori knowledge of prior or encoding. Simulations and applications on the basis of behavioral datasets validate that the predictions of this theory apply in realistic settings. Importantly, our results demonstrate that reliable recovery of the model often requires having data from multiple noise levels. This is a crucial insight that will guide future experimental design.
Topological structure in human spatial representation revealed through drawing Human spatial representations are often assumed to represent Euclidean properties such as length, distance, and angle. Here we test an alternative (but not mutually exclusive) possibility – that spatial memory is structured primarily around topological relations. Across four experiments, adults and children memorized simple letter-like figures and reproduced them by drawing, allowing the contents of their spatial representations to be revealed directly. Drawings showed systematic distortions of metric features, including strong biases of angles toward 90° and compression of line length towards an average value. In contrast, topologically critical features — such as T-junctions and holes — were reliably preserved, even relative to closely matched but topologically irrelevant features like L-junctions. These effects were magnified in a serial reproduction paradigm, in which participants iteratively generated new drawings from previous participant drawings: At the end of each mnemonic chain, figures converged on simplified topological structures as metric detail degraded. Similar patterns were observed in children aged five to eight years. Together, these findings suggest that basic topological relations may function as primitive building blocks of human spatial representation, with metric detail encoded secondarily.
Neural traveling waves in cortex: Network mechanisms and potential roles in neural computation First discovered in anesthetized animals, neural traveling waves (nTWs) have now been observed throughout the brains of awake animals, where they influence neural excitability and behavior. nTWs can arise intrinsically in ongoing network activity or be triggered by sensory input and behavioral events. By structuring neural activity within individual cortical regions, nTWs introduce spatiotemporal dependencies across sensory maps that are not naturally captured by purely feedforward or feedback processing. Here, we synthesize physiological and computational evidence around two themes, focusing on the visual system while drawing connections to other cortical areas. First, we define nTWs and outline the circuit mechanisms that can generate them. Second, we highlight how nTWs can implement spatiotemporal computations, such as predicting upcoming sensory inputs, by embedding sensory history in the evolving activity pattern of an individual cortical region. We conclude by introducing a conceptual framework for spatiotemporal, generative processing by nTWs traveling over sensory maps.
Age and Physical Activity Modulate the Spatial Mapping of Time-Related Words The processing of temporal concepts is known to be intertwined with spatial cognition. For instance, reaction times (RTs) are shorter when participants classify past- and future-related words with left- and right-lateral responses, respectively, in line with the so-called mental time line hypothesis. While some research has addressed individual differences in processing temporal concepts, it still remains unclear how such space–time congruency effects are influenced by the most crucial time-related factor in our lives – age. To address this question, we asked 94 participants aged 19 to 80 years old to classify past, future, and neutral words as related to time by using left and right response keys. In addition, we collected information regarding participants’ self-reported level of physical activity, as it was shown to be a predictor of healthy ageing. We found that younger participants processed future-related words faster than past-related words, whereas older participants processed past-related words faster than future-related words. This ageing effect was modulated by physical exercise: The higher the self-reported level of physical activity in older participants, the smaller the past-related RT bias. Finally, our data revealed an age-related decline in space–time congruency effects among participants with lower self-reported physical activity. These findings indicate that the conceptual representation of time undergoes lifelong changes influenced by multiple factors, including age and physical activity. We suggest an integrative construct of dynamic conceptual time whereby the individual’s bodily and life-course experiences lead to gradual changes in the processing of temporal concepts.
The Cortical Contribution to the Speech-Frequency–Following Response Is Not Modulated by Visual Information Seeing a speaker’s face can significantly aid understanding, particularly in challenging acoustic environments. An early neural response implicated in audiovisual speech processing is the frequency-following response (speech-FFR), which occurs at the fundamental frequency of the speech signal. This response arises from both subcortical areas and the auditory cortex. A previous study showed that subcortical responses to speech can be enhanced when a listener can see the talker’s face. Here, we examined the cortical contribution to the speech-FFR and its potential modulation by visual information, motivated in part by the growing use of artificially generated talking-face avatars to support speech comprehension. We recorded MEG responses to four types of audiovisual signals: a still image, an artificially generated avatar, a degraded video, and a natural video. The audio stimuli were presented in a substantial level of background noise to make behavioral audiovisual effects stand out. Speech-in-noise comprehension increased significantly from the audio-only condition to the avatar and the degraded video, and further to the natural video. Moreover, we found that all types of audiovisual stimuli yielded robust speech-FFRs in the auditory cortex at an early latency of around 30 ms. However, the magnitude of this neural response was neither enhanced nor attenuated by the videos, nor could the cortical contribution of the speech-FFR explain a significant portion of the variance in the behavioral comprehension scores. Our results suggest that visual modulation of the speech-FFR in the auditory cortex is, if existent, too small to be measurable in scenarios where speech occurs in considerable background noise.