Abstract Perceptions are the result of constructivist processes. The brain has to extract from the stream of sensory activity those spatiotemporal patterns that signal the presence of informative features, associate those features with one another that belong to individual perceptual objects, segregate these objects from one another and the background, identify the nature of the objects, examine the relations between them, and provide a meaningful interpretation of the respective constellation. These operations require the evaluation of spatial and temporal relations between neuronal signals and subsequently the selective association of signals in a meaningful way. The question how these selective associations of features are achieved is addressed as the “binding problem.” It requires for its solution that the brain has a priori knowledge about the probability that certain features are more likely related with one another than others. Thus, the brain needs to have an internal model of the statistical regularities of the world from which it can infer the rules according to which features should be associated. These rules, addressed as the Gestalt rules of perception, reside in the functional architecture of the brain’s sensory systems. The basic layout of this architecture is determined genetically and shows little variation among members of the same species. However, in higher vertebrates and all mammals, the functional architecture of sensory systems is highly susceptible to experience-dependent modifications during critical periods of early development. And even in the mature brain, the synaptic connections in sensory stems remain plastic, allowing for lifelong perceptual learning. Hence, the internal model of the world is based on knowledge acquired during both evolution and postnatal life. The challenge is to identify the mechanisms that permit the exploitation of this knowledge to solve binding problems. Since the 1990s, several mechanisms have been proposed, and it is likely that they coexist and complement one another. One well-established strategy is the implementation of conjunction specific neurons. Afferents signaling the presence of features that are to be bound are made to converge on common target cells. With appropriate adjustments of thresholds, the responses of these cells would then signal a particular constellation of features. This strategy can be realized in simple feedforward architectures but is expensive in terms of hardware because every constellation of features requires for its binding a devoted anatomical circuit. Therefore, additional strategies are exploited that allow for a more flexible, dynamic association of features. One proposal is that features are bound by allocation of selective attention to specific feature constellations. Another proposal is that specific feature constellations are represented by dynamically formed assemblies of distributed feature selective neurons. These assemblies are functionally coherent entities that require for their formation reciprocal interactions and therefore can only be realized in recurrent networks. This binding strategy economizes on hardware because different constellations of feature-specific neurons can be flexibly bound into assemblies. However, this binding strategy requires signatures that identify those feature-specific neurons that are temporarily bound into a functionally coherent assembly. Candidates for such signatures are joint rate increases or enhanced synchrony among discharges of neurons having joined a particular assembly or both. Both phenomena have been observed. If joint rate increases were the only signature of relatedness, assemblies representing different but spatially contiguous objects might become confounded. This superposition problem is alleviated if the signature of relatedness is complemented by precise synchronization of discharges because assemblies can then be segregated by temporal offsets or, if responses are oscillatory, in phase space. The readout of binding results differs for the different binding strategies. In case of binding through anatomical convergence, the binding result is signaled by the activity of conjunction-specific neurons. In case of dynamic binding, the spatiotemporally structured activity patterns of assemblies need to be conveyed in parallel to recurrent networks in downstream areas where they can then ignite again assemblies according to the priors stored in the architecture of these areas. In this way, areas occupying different levels of the processing hierarchy can perform binding operations at different spatial and temporal scales. Because of reciprocal coupling, binding solutions achieved at different levels of the processing hierarchy can be compared, allowing convergence toward dynamic states that represent the best match between sensory evidence and stored priors. These distributed dynamic states could be mapped directly onto executive areas such as the motor cortex and serve the orchestration of the activity vectors required for the control of movements.

Oscillatory recurrent networks, such as the harmonic oscillator recurrent network (HORN) model, offer advantages in parameter efficiency, learning speed, and robustness relative to traditional nonoscillating architectures. Yet, while many implementations of physical neural networks exploiting attractor dynamics have been studied, so far implementations of oscillatory models in analog-electronic hardware that utilize the networks’ transient dynamics are lacking. This study explores the feasibility of implementing HORNs in analog-electronic hardware while maintaining the computational performance of the digital counterpart. Using a digital-twin approach, we trained a four-node HORN for sMNIST (sequential Modified National Institute of Standards and Technology) classification and transferred the trained parameters to an analog-electronic implementation. A set of custom error metrics indicated that the analog system is able to successfully replicate the dynamics of the digital model in most test cases. However, despite the overall well-matching dynamics, when using the readout layer of the digital model on the data generated by the analog system, we observed only 28.39% agreement with the predictions of the digital model. An analysis shows that this mismatch is due to a precision difference between the analog hardware and the floating-point representation exploited by the digital model to perform classification tasks. When the analog system was utilized as a reservoir with a retrained linear readout, its classification performance could be recovered to that of the digital twin, indicating preserved information content within the analog dynamics. This proof of concept establishes that analog-electronic circuits can effectively implement oscillatory neural networks for computation, providing a demonstration of energy-efficient analog systems that exploit brain-inspired transient dynamics for computation.

Several areas of cognitive neuroscience tackle traditional philosophical questions. Among the range of problems, two closely related issues will be addressed in more detail from both neurobiological and philosophical perspectives: the relationship between mind and matter and the nature of perception. Neuropsychological and neurophysiological studies are reviewed that examine the connection between neuronal processes and consciousness. The most prominent theories on the neuronal correlates of consciousness (NCC) are then compared with philosophical attempts to address the epistemic gap between the material processes in the brain and mental phenomena. Before exploring whether neurobiological discoveries can help resolve philosophical problems, the epistemic challenges are discussed, stemming from the fact that perceptions are shaped by the brain's functional architecture. It is suggested that the 'hard problem of consciousness'-the challenge of explaining how the qualia of subjective experience can arise from neuronal processes-can be alleviated if two conditions are met: first, that perception depends on priors and, second, that some of these priors are formed through interactions with the immaterial realities of cultural concepts. Although this approach offers a coherent naturalistic explanation for the emergence of mental phenomena, it does not resolve the cognitive dissonance between our intuitions and scientific evidence regarding the relationship between matter and mind.

The first part of the paper is devoted to a comparison between the functional architectures of the cerebral cortex and artificial intelligent systems. While the two systems share numerous features, natural systems differ in at least four important aspects: i) the prevalence of recurrent connections, ii) the ability to use the temporal domain for computations, iii) the ability to perform "in memory" computations and iv) the prevalence of analog computations. The second part of the paper focuses on a simulation study that has been designed to answer the long-standing question of whether the oscillatory patterning of neuronal activity, which is a hallmark of natural systems, is an epiphenomenon of recurrent interactions or serves a functional role. To this end, recurrent neuronal networks were simulated to capture essential features of cortical networks, and their performance was tested on standard pattern recognition benchmark tests. In order to control the oscillatory regime of these networks, its nodes were configured as damped harmonic oscillators. By varying the damping factor, the nodes functioned either as leaky integrators or oscillators. It turned out that networks with oscillatory nodes substantially outperformed their non-oscillating counterparts. The reasons for this superior performance and similarities with natural neuronal networks are discussed. It is concluded that the oscillatory patterning of neuronal responses is functionally relevant because it allows the exploitation of the unique dynamics of coupled oscillators for analog computation.

The dynamics of neuronal systems are characterized by hallmark features such as oscillations and synchrony. However, it has remained unclear whether these characteristics are epiphenomena or are exploited for computation. Due to the challenge of selectively interfering with oscillatory network dynamics in neuronal systems, we simulated recurrent networks of damped harmonic oscillators in which oscillatory activity is enforced in each node, a choice well supported by experimental findings. When trained on standard pattern recognition tasks, these harmonic oscillator recurrent networks (HORNs) outperformed nonoscillatory architectures with respect to learning speed, noise tolerance, and parameter efficiency. HORNs also reproduced a many characteristic features of neuronal systems, such as the cerebral cortex and the hippocampus. In trained HORNs, stimulus-induced interference patterns holistically represent the result of comparing sensory evidence with priors stored in recurrent connection weights, and learning-induced weight changes are compatible with Hebbian principles. Implementing additional features characteristic of natural networks, such as heterogeneous oscillation frequencies, inhomogeneous conduction delays, and network modularity, further enhanced HORN performance without requiring additional parameters. Taken together, our model allows us to give plausible a posteriori explanations for features of natural networks whose computational role has remained elusive. We conclude that neuronal systems are likely to exploit the unique dynamics of recurrent oscillator networks whose computational superiority critically depends on the oscillatory patterning of their nodal dynamics. Implementing the proposed computational principles in analog hardware is expected to enable the design of highly energy-efficient and self-adapting devices that could ideally complement existing digital technologies.

Natural scene responses in the primary visual cortex are modulated simultaneously by attention and by contextual signals about scene statistics stored across the connectivity of the visual processing hierarchy. We hypothesized that attentional and contextual signals interact in V1 in a manner that primarily benefits the representation of natural stimuli, rich in high -order statistical structure. Recording from two macaques engaged in a spatial attention task, we found that attention enhanced the decodability of stimulus identity from population responses evoked by natural scenes, but not by synthetic stimuli lacking higher -order statistical regularities. Population analysis revealed that neuronal responses converged to a low -dimensional subspace only for natural stimuli. Critically, we determined that the attentional enhancement in stimulus decodability was captured by the natural -scene subspace, indicating an alignment between the attentional and natural stimulus variance. These results suggest that attentional and contextual signals interact in V1 in a manner optimized for natural vision.

We investigated whether neurons in monkey primary visual cortex (V1) exhibit mixed selectivity for sensory input and behavioral choice. Parallel multisite spiking activity was recorded from area V1 of awake monkeys performing a delayed match-to-sample task. The monkeys had to make a forced choice decision of whether the test stimulus matched the preceding sample stimulus. The population responses evoked by the test stimulus contained information about both the identity of the stimulus and with some delay but before the onset of the motor response the forthcoming choice. The results of subspace identification analysis indicate that stimulus-specific and decision-related information coexists in separate subspaces of the high-dimensional population activity, and latency considerations suggest that the decision-related information is conveyed by top-down projections.

In order to investigate the involvement of the primary visual cortex (V1) in working memory (WM), parallel, multisite recordings of multi-unit activity were obtained from monkey V1 while the animals performed a delayed match-to-sample (DMS) task. During the delay period, V1 population firing rate vectors maintained a lingering trace of the sample stimulus that could be reactivated by intervening impulse stimuli that enhanced neuronal firing. This fading trace of the sample did not require active engagement of the monkeys in the DMS task and likely reflects the intrinsic dynamics of recurrent cortical networks in lower visual areas. This renders an active, attention-dependent involvement of V1 in the maintenance of WM contents unlikely. By contrast, population responses to the test stimulus depended on the probabilistic contingencies between sample and test stimuli. Responses to tests that matched expectations were reduced which agrees with concepts of predictive coding.

BACKGROUND:Reduced auditory mismatch negativity (MMN) is robustly impaired in schizophrenia. However, mechanisms underlying dysfunctional MMN generation remain incompletely understood. This study aimed to examine the role of evoked spectral power and phase-coherence towards deviance detection and its impairments in schizophrenia.METHODS:Magnetoencephalography data was collected in 16 male schizophrenia patients and 16 male control participants during an auditory MMN paradigm. Analyses of event-related fields (ERF), spectral power and inter-trial phase-coherence (ITPC) focused on Heschl's gyrus, superior temporal gyrus, inferior/medial frontal gyrus and thalamus.RESULTS:MMNm ERF amplitudes were reduced in patients in temporal, frontal and subcortical regions, accompanied by decreased theta-band responses, as well as by a diminished gamma-band response in auditory cortex. At theta/alpha frequencies, ITPC to deviant tones was reduced in patients in frontal cortex and thalamus. Patients were also characterized by aberrant responses to standard tones as indexed by reduced theta-/alpha-band power and ITPC in temporal and frontal regions. Moreover, stimulus-specific adaptation was decreased at theta/alpha frequencies in left temporal regions, which correlated with reduced MMNm spectral power and ERF amplitude. Finally, phase-reset of alpha-oscillations after deviant tones in left thalamus was impaired, which correlated with impaired MMNm generation in auditory cortex. Importantly, both non-rhythmic and rhythmic components of spectral activity contributed to the MMNm response.CONCLUSIONS:Our data indicate that deficits in theta-/alpha- and gamma-band activity in cortical and subcortical regions as well as impaired spectral responses to standard sounds could constitute potential mechanisms for dysfunctional MMN generation in schizophrenia, providing a novel perspective towards MMN deficits in the disorder.

Dr. Wolf Singer (b. 1943) is one of Germany's most renowned brain researchers and neurophysiologists. His accomplishments in the creation of new research centers for neuroscience as well as his commitment to European scientific organizations for integrative brain research are highly valued as significant moments of advancement in the neurosciences. Before his appointment as a scientific member of the Max Planck Society and director at the Frankfurt Max Planck Institute for Brain Research, he gained deep insight into the chances and pitfalls of translational initiatives at the Max Planck Institute of Psychiatry in Munich. From the late 1950s onward, the institute adapted to emerging international trends and successfully integrated neurochemistry, neurophysiology, and neuroanatomy into the fledgling interdisciplinary field of neuroscience. This agenda of reorientation was an undertaking of Otto Detlev Creutzfeldt, Detlev Ploog, Gerd Peters, and Horst Jatzkewitz, among others. In the 1970s, Munich's laboratories attracted scientists from several countries in Europe and abroad. This article examines whether specific styles of conducting (neuro)science research existed in the Max Planck Society.

Perceptual decision making is a fundamentally hierarchical inference task consisting of two steps. In the first perceptual step, stimulus identity needs to be inferred. In the second step, given the inferred stimulus identity, the correct decision needs to be inferred. At which of these levels are prior expectations used to inform inferences? Previous approaches studying the role of priors in perceptual decision making did not distinguish between these two levels, and focused almost exclusively on the decision level. Here, combining approaches from non-human primate and human infant research, we developed a novel experimental paradigm that allowed us to identify behavioral signatures of perceptual versus decision priors. We trained two macaque monkeys in a modified variant of a delayed match-to-sample task in which we used three stimuli with carefully controlled statistics. The prior over match vs. non-match decisions was uniform, but (irrelevant to the match/non-match decision) each sample stimulus could be followed by one of two different non-match target stimuli drawn from a highly skewed prior. Thus, the prior over target stimuli changed dynamically, trial-by-trial, conditioned on the sample stimulus. In standard decision trials, monkeys were required to maintain central fixation while stimuli were presented peripherally and had to make a match/non-match decision by controlling a lever. Occasional free-viewing trials used an identical design, except that no decision was required and monkeys were allowed to break fixation on the presentation of the target, while we tracked their eye movements. Both accuracy in decision trials, and conditional familiarity as expressed by differential looking times in free-viewing trials showed significant differences between common and rare non-match targets. These results provide converging evidence from two different response modalities for the internal representation of dynamically changing task-irrelevant perceptual priors, opening the way to the study of their neural underpinning.

The Eureka effect refers to the common experience of suddenly solving a problem. Here, we study this effect in a pattern recognition paradigm that requires the segmentation of complex scenes and recognition of objects on the basis of Gestalt rules and prior knowledge. In the experiments, both sensory evidence and prior knowledge were manipulated in order to obtain trials that do or do not converge toward a perceptual solution. Subjects had to detect objects in blurred scenes and indicate recognition with manual responses. Neural dynamics were assessed with high-density Electroencephalography (EEG) recordings. The results show significant changes of neural dynamics with respect to spectral distribution, coherence, phase locking, and fractal dimensionality. The Eureka effect was associated with increased coherence of oscillations in the alpha and theta bands over widely distributed regions of the cortical mantle predominantly in the right hemisphere. This increase in coherence was associated with decreased beta power over parietal and central regions and with decreased alpha power over frontal and occipital areas. In addition, there was a right hemisphere-lateralized reduction of fractal dimensionality. We propose that the Eureka effect requires cooperation of cortical regions involved in working memory, creative thinking, and the control of attention.

Wolf Singer has long been interested in neural dynamics, synchrony, and temporal codes. On his 80th birthday, he talks with Neuron about his seminal contributions, the need to engage with the public on philosophical and ethical implications of scientific research, and further speculation about the future of neuroscience.

Parallel multisite recordings in the visual cortex of trained monkeys revealed that the responses of spatially distributed neurons to natural scenes are ordered in sequences. The rank order of these sequences is stimulus-specific and maintained even if the absolute timing of the responses is modified by manipulating stimulus parameters. The stimulus specificity of these sequences was highest when they were evoked by natural stimuli and deteriorated for stimulus versions in which certain statistical regularities were removed. This suggests that the response sequences result from a matching operation between sensory evidence and priors stored in the cortical network. Decoders trained on sequence order performed as well as decoders trained on rate vectors but the former could decode stimulus identity from considerably shorter response intervals than the latter. A simulated recurrent network reproduced similarly structured stimulus-specific response sequences, particularly once it was familiarized with the stimuli through non-supervised Hebbian learning. We propose that recurrent processing transforms signals from stationary visual scenes into sequential responses whose rank order is the result of a Bayesian matching operation. If this temporal code were used by the visual system it would allow for ultrafast processing of visual scenes.

Predictive coding is an important candidate theory of self-supervised learning in the brain. Its central idea is that sensory responses result from comparisons between bottom-up inputs and contextual predictions, a process in which rates and synchronization may play distinct roles. We recorded from awake macaque V1 and developed a technique to quantify stimulus predictability for natural images based on self-supervised, generative neural networks. We find that neuronal firing rates were mainly modulated by the contextual predictability of higher-order image features, which correlated strongly with human perceptual similarity judgments. By contrast, V1 gamma (γ)-synchronization increased monotonically with the contextual predictability of low-level image features and emerged exclusively for larger stimuli. Consequently, γ-synchronization was induced by natural images that are highly compressible and low-dimensional. Natural stimuli with low predictability induced prominent, late-onset beta (β)-synchronization, likely reflecting cortical feedback. Our findings reveal distinct roles of synchronization and firing rates in the predictive coding of natural images.

Individual differences in perception are widespread. Considering inter-individual variability, synesthetes experience stable additional sensations; schizophrenia patients suffer perceptual deficits in, eg, perceptual organization (alongside hallucinations and delusions). Is there a unifying principle explaining inter-individual variability in perception? There is good reason to believe perceptual experience results from inferential processes whereby sensory evidence is weighted by prior knowledge about the world. Perceptual variability may result from different precision weighting of sensory evidence and prior knowledge. We tested this hypothesis by comparing visibility thresholds in a perceptual hysteresis task across medicated schizophrenia patients (N = 20), synesthetes (N = 20), and controls (N = 26). Participants rated the subjective visibility of stimuli embedded in noise while we parametrically manipulated the availability of sensory evidence. Additionally, precise long-term priors in synesthetes were leveraged by presenting either synesthesia-inducing or neutral stimuli. Schizophrenia patients showed increased visibility thresholds, consistent with overreliance on sensory evidence. In contrast, synesthetes exhibited lowered thresholds exclusively for synesthesia-inducing stimuli suggesting high-precision long-term priors. Additionally, in both synesthetes and schizophrenia patients explicit, short-term priors-introduced during the hysteresis experiment-lowered thresholds but did not normalize perception. Our results imply that perceptual variability might result from differences in the precision afforded to prior beliefs and sensory evidence, respectively.

Introduction Illuminating neurobiological mechanisms underlying the protective effect of recently discovered common genetic resilience variants for schizophrenia is crucial for more effective prevention efforts. Current models implicate adaptive neuroplastic changes in the visual system and their pro-cognitive effects as a schizophrenia resilience mechanism. We investigated whether common genetic resilience variants might affect brain structure in similar neural circuits. Method Using structural magnetic resonance imaging, we measured the impact of an established schizophrenia polygenic resilience score (PRSResilience) on cortical volume, thickness, and surface area in 101 healthy subjects and in a replication sample of 33 224 healthy subjects (UK Biobank). Finding We observed a significant positive whole-brain correlation between PRSResilience and cortical volume in the right fusiform gyrus (FFG) (r = 0.35; P = .0004). Post-hoc analyses in this cluster revealed an impact of PRSResilience on cortical surface area. The replication sample showed a positive correlation between PRSResilience and global cortical volume and surface area in the left FFG. Conclusion Our findings represent the first evidence of a neurobiological correlate of a genetic resilience factor for schizophrenia. They support the view that schizophrenia resilience emerges from strengthening neural circuits in the ventral visual pathway and an increased capacity for the disambiguation of social and nonsocial visual information. This may aid psychosocial functioning, ameliorate the detrimental effects of subtle perceptual and cognitive disturbances in at-risk individuals, and facilitate coping with the cognitive and psychosocial consequences of stressors. Our results thus provide a novel link between visual cognition, the vulnerability-stress concept, and schizophrenia resilience models.