Learning to Estimate Dynamical State with Probabilistic Population Codes

Tracking moving objects, including one’s own body, is a fundamental ability of higher organisms, playing a central role in many perceptual and motor tasks. While it is unknown how the brain learns to follow and predict the dynamics of objects, it is known that this process of state estimation can be learned purely from the statistics of noisy observations. When the dynamics are simply linear with additive Gaussian noise, the optimal solution is the well known Kalman filter (KF), the parameters of which can be learned via latent-variable density estimation (the EM algorithm). The brain does not, however, directly manipulate matrices and vectors, but instead appears to represent probability distributions with the firing rates of population of neurons, “probabilistic population codes.” We show that a recurrent neural network—a modified form of an exponential family harmonium (EFH)—that takes a linear probabilistic population code as input can learn, without supervision, to estimate the state of a linear dynamical system. After observing a series of population responses (spike counts) to the position of a moving object, the network learns to represent the velocity of the object and forms nearly optimal predictions about the position at the next time-step. This result builds on our previous work showing that a similar network can learn to perform multisensory integration and coordinate transformations for static stimuli. The receptive fields of the trained network also make qualitative predictions about the developing and learning brain: tuning gradually emerges for higher-order dynamical states not explicitly present in the inputs, appearing as delayed tuning for the lower-order states.     

Sustained neural rhythms reveal endogenous oscillations supporting speech perception

Rhythmic sensory or electrical stimulation will produce rhythmic brain responses. These rhythmic responses are often interpreted as endogenous neural oscillations aligned (or “entrained”) to the stimulus rhythm. However, stimulus-aligned brain responses can also be explained as a sequence of evoked responses, which only appear regular due to the rhythmicity of the stimulus, without necessarily involving underlying neural oscillations. To distinguish evoked responses from true oscillatory activity, we tested whether rhythmic stimulation produces oscillatory responses which continue after the end of the stimulus. Such sustained effects provide evidence for true involvement of neural oscillations. In Experiment 1, we found that rhythmic intelligible, but not unintelligible speech produces oscillatory responses in magnetoencephalography (MEG) which outlast the stimulus at parietal sensors. In Experiment 2, we found that transcranial alternating current stimulation (tACS) leads to rhythmic fluctuations in speech perception outcomes after the end of electrical stimulation. We further report that the phase relation between electroencephalography (EEG) responses and rhythmic intelligible speech can predict the tACS phase that leads to most accurate speech perception. Together, we provide fundamental results for several lines of research—including neural entrainment and tACS—and reveal endogenous neural oscillations as a key underlying principle for speech perception.

Just as a child on a swing continues to move after the pushing stops, this study reveals similar entrained rhythmic echoes in brain activity after hearing speech and electrical brain stimulation; perturbation with tACS shows that these brain oscillations help listeners to understand speech.     

A Generalized Framework for Quantifying the Dynamics of EEG Event-Related Desynchronization

Brains were built by evolution to react swiftly to environmental challenges. Thus, sensory stimuli must be processed ad hoc, i.e., independent—to a large extent—from the momentary brain state incidentally prevailing during stimulus occurrence. Accordingly, computational neuroscience strives to model the robust processing of stimuli in the presence of dynamical cortical states. A pivotal feature of ongoing brain activity is the regional predominance of EEG eigenrhythms, such as the occipital alpha or the pericentral mu rhythm, both peaking spectrally at 10 Hz. Here, we establish a novel generalized concept to measure event-related desynchronization (ERD), which allows one to model neural oscillatory dynamics also in the presence of dynamical cortical states. Specifically, we demonstrate that a somatosensory stimulus causes a stereotypic sequence of first an ERD and then an ensuing amplitude overshoot (event-related synchronization), which at a dynamical cortical state becomes evident only if the natural relaxation dynamics of unperturbed EEG rhythms is utilized as reference dynamics. Moreover, this computational approach also encompasses the more general notion of a “conditional ERD,” through which candidate explanatory variables can be scrutinized with regard to their possible impact on a particular oscillatory dynamics under study. Thus, the generalized ERD represents a powerful novel analysis tool for extending our understanding of inter-trial variability of evoked responses and therefore the robust processing of environmental stimuli.     

Stimulus Set Meaningfulness and Neurophysiological Differentiation: A Functional Magnetic Resonance Imaging Study

A meaningful set of stimuli, such as a sequence of frames from a movie, triggers a set of different experiences. By contrast, a meaningless set of stimuli, such as a sequence of ‘TV noise’ frames, triggers always the same experience—of seeing ‘TV noise’—even though the stimuli themselves are as different from each other as the movie frames. We reasoned that the differentiation of cortical responses underlying the subject’s experiences, as measured by Lempel-Ziv complexity (incompressibility) of functional MRI images, should reflect the overall meaningfulness of a set of stimuli for the subject, rather than differences among the stimuli. We tested this hypothesis by quantifying the differentiation of brain activity patterns in response to a movie sequence, to the same movie scrambled in time, and to ‘TV noise’, where the pixels from each movie frame were scrambled in space. While overall cortical activation was strong and widespread in all conditions, the differentiation (Lempel-Ziv complexity) of brain activation patterns was correlated with the meaningfulness of the stimulus set, being highest in the movie condition, intermediate in the scrambled movie condition, and minimal for ‘TV noise’. Stimulus set meaningfulness was also associated with higher information integration among cortical regions. These results suggest that the differentiation of neural responses can be used to assess the meaningfulness of a given set of stimuli for a given subject, without the need to identify the features and categories that are relevant to the subject, nor the precise location of selective neural responses.     

Why is Real-World Visual Object Recognition Hard?

Progress in understanding the brain mechanisms underlying vision requires the construction of computational models that not only emulate the brain's anatomy and physiology, but ultimately match its performance on visual tasks. In recent years, “natural” images have become popular in the study of vision and have been used to show apparently impressive progress in building such models. Here, we challenge the use of uncontrolled “natural” images in guiding that progress. In particular, we show that a simple V1-like model—a neuroscientist's “null” model, which should perform poorly at real-world visual object recognition tasks—outperforms state-of-the-art object recognition systems (biologically inspired and otherwise) on a standard, ostensibly natural image recognition test. As a counterpoint, we designed a “simpler” recognition test to better span the real-world variation in object pose, position, and scale, and we show that this test correctly exposes the inadequacy of the V1-like model. Taken together, these results demonstrate that tests based on uncontrolled natural images can be seriously misleading, potentially guiding progress in the wrong direction. Instead, we reexamine what it means for images to be natural and argue for a renewed focus on the core problem of object recognition—real-world image variation.     

What are lightness illusions and why do we see them?

Lightness illusions are fundamental to human perception, and yet why we see them is still the focus of much research. Here we address the question by modelling not human physiology or perception directly as is typically the case but our natural visual world and the need for robust behaviour. Artificial neural networks were trained to predict the reflectance of surfaces in a synthetic ecology consisting of 3-D “dead-leaves” scenes under non-uniform illumination. The networks learned to solve this task accurately and robustly given only ambiguous sense data. In addition—and as a direct consequence of their experience—the networks also made systematic “errors” in their behaviour commensurate with human illusions, which includes brightness contrast and assimilation—although assimilation (specifically White's illusion) only emerged when the virtual ecology included 3-D, as opposed to 2-D scenes. Subtle variations in these illusions, also found in human perception, were observed, such as the asymmetry of brightness contrast. These data suggest that “illusions” arise in humans because (i) natural stimuli are ambiguous, and (ii) this ambiguity is resolved empirically by encoding the statistical relationship between images and scenes in past visual experience. Since resolving stimulus ambiguity is a challenge faced by all visual systems, a corollary of these findings is that human illusions must be experienced by all visual animals regardless of their particular neural machinery. The data also provide a more formal definition of illusion: the condition in which the true source of a stimulus differs from what is its most likely (and thus perceived) source. As such, illusions are not fundamentally different from non-illusory percepts, all being direct manifestations of the statistical relationship between images and scenes.     

What Is Stochastic Resonance? Definitions, Misconceptions, Debates, and Its Relevance to Biology

Stochastic resonance is said to be observed when increases in levels of unpredictable fluctuations—e.g., random noise—cause an increase in a metric of the quality of signal transmission or detection performance, rather than a decrease. This counterintuitive effect relies on system nonlinearities and on some parameter ranges being “suboptimal”. Stochastic resonance has been observed, quantified, and described in a plethora of physical and biological systems, including neurons. Being a topic of widespread multidisciplinary interest, the definition of stochastic resonance has evolved significantly over the last decade or so, leading to a number of debates, misunderstandings, and controversies. Perhaps the most important debate is whether the brain has evolved to utilize random noise in vivo, as part of the “neural code”. Surprisingly, this debate has been for the most part ignored by neuroscientists, despite much indirect evidence of a positive role for noise in the brain. We explore some of the reasons for this and argue why it would be more surprising if the brain did not exploit randomness provided by noise—via stochastic resonance or otherwise—than if it did. We also challenge neuroscientists and biologists, both computational and experimental, to embrace a very broad definition of stochastic resonance in terms of signal-processing “noise benefits”, and to devise experiments aimed at verifying that random variability can play a functional role in the brain, nervous system, or other areas of biology.     

On the Relevance of Natural Stimuli for the Study of Brainstem Correlates: The Example of Consonance Perception

Some combinations of musical tones sound pleasing to Western listeners, and are termed consonant, while others sound discordant, and are termed dissonant. The perceptual phenomenon of consonance has been traced to the acoustic property of harmonicity. It has been repeatedly shown that neural correlates of consonance can be found as early as the auditory brainstem as reflected in the harmonicity of the scalp-recorded frequency-following response (FFR). “Neural Pitch Salience” (NPS) measured from FFRs—essentially a time-domain equivalent of the classic pattern recognition models of pitch—has been found to correlate with behavioral judgments of consonance for synthetic stimuli. Following the idea that the auditory system has evolved to process behaviorally relevant natural sounds, and in order to test the generalizability of this finding made with synthetic tones, we recorded FFRs for consonant and dissonant intervals composed of synthetic and natural stimuli. We found that NPS correlated with behavioral judgments of consonance and dissonance for synthetic but not for naturalistic sounds. These results suggest that while some form of harmonicity can be computed from the auditory brainstem response, the general percept of consonance and dissonance is not captured by this measure. It might either be represented in the brainstem in a different code (such as place code) or arise at higher levels of the auditory pathway. Our findings further illustrate the importance of using natural sounds, as a complementary tool to fully-controlled synthetic sounds, when probing auditory perception.     

Phenotypic flexibility as key factor in the human nutrition and health relationship

Metabolic adaptation to a disturbance of homeostasis is determined by a series of interconnected physiological processes and molecular mechanisms that can be followed in space (i.e., different organs or organelles) and in time. The amplitudes of these responses of this “systems flexibility network” determine to what extent the individual can adequately react to external challenges of varying nature and thus determine the individual’s health status and disease predisposition. Connected pathways and regulatory networks act as “adaptive response systems” with metabolic and inflammatory processes as a core—but embedded into psycho-neuro-endocrine control mechanisms that in their totality define the phenotypic flexibility in an individual. Optimal metabolic health is thus the orchestration of all mechanisms and processes that maintain this flexibility in an organism as a phenotype. Consequently, onset of many chronic metabolic diseases results from impairment or even loss of flexibility in parts of the system. This also means that metabolic diseases need to be diagnosed and treated from a systems perspective referring to a “systems medicine” approach. This requires a far better understanding of the mechanisms involved in maintaining, optimizing and restoring phenotypic flexibility. Although a loss of flexibility in a specific part of the network may promote pathologies, this not necessarily takes place in the same part because the system compensates. Diagnosis at systems level therefore needs the quantification of the response reactions of all relevant parts of the phenotypic flexibility system. This can be achieved by disturbing the homeostatic system by any challenge from extended fasting, to intensive exercise or a caloric overload.     

Cell Lineages and the Logic of Proliferative Control

It is widely accepted that the growth and regeneration of tissues and organs is tightly controlled. Although experimental studies are beginning to reveal molecular mechanisms underlying such control, there is still very little known about the control strategies themselves. Here, we consider how secreted negative feedback factors (“chalones”) may be used to control the output of multistage cell lineages, as exemplified by the actions of GDF11 and activin in a self-renewing neural tissue, the mammalian olfactory epithelium (OE). We begin by specifying performance objectives—what, precisely, is being controlled, and to what degree—and go on to calculate how well different types of feedback configurations, feedback sensitivities, and tissue architectures achieve control. Ultimately, we show that many features of the OE—the number of feedback loops, the cellular processes targeted by feedback, even the location of progenitor cells within the tissue—fit with expectations for the best possible control. In so doing, we also show that certain distinctions that are commonly drawn among cells and molecules—such as whether a cell is a stem cell or transit-amplifying cell, or whether a molecule is a growth inhibitor or stimulator—may be the consequences of control, and not a reflection of intrinsic differences in cellular or molecular character.     

Focus: The Aging Brain: Neural stem/progenitor cells in Alzheimer’s disease

Alzheimer’s disease (AD) is the most prevalent neurodegenerative disease and a worldwide health challenge. Different therapeutic approaches are being developed to reverse or slow the loss of affected neurons. Another plausible therapeutic way that may complement the studies is to increase the survival of existing neurons by mobilizing the existing neural stem/progenitor cells (NSPCs) — i.e. “induce their plasticity” — to regenerate lost neurons despite the existing pathology and unfavorable environment. However, there is controversy about how NSPCs are affected by the unfavorable toxic environment during AD. In this review, we will discuss the use of stem cells in neurodegenerative diseases and in particular how NSPCs affect the AD pathology and how neurodegeneration affects NSPCs. In the end of this review, we will discuss how zebrafish as a useful model organism with extensive regenerative ability in the brain might help to address the molecular programs needed for NSPCs to respond to neurodegeneration by enhanced neurogenesis.     

Associative Processing Is Inherent in Scene Perception

How are complex visual entities such as scenes represented in the human brain? More concretely, along what visual and semantic dimensions are scenes encoded in memory? One hypothesis is that global spatial properties provide a basis for categorizing the neural response patterns arising from scenes. In contrast, non-spatial properties, such as single objects, also account for variance in neural responses. The list of critical scene dimensions has continued to grow—sometimes in a contradictory manner—coming to encompass properties such as geometric layout, big/small, crowded/sparse, and three-dimensionality. We demonstrate that these dimensions may be better understood within the more general framework of associative properties. That is, across both the perceptual and semantic domains, features of scene representations are related to one another through learned associations. Critically, the components of such associations are consistent with the dimensions that are typically invoked to account for scene understanding and its neural bases. Using fMRI, we show that non-scene stimuli displaying novel associations across identities or locations recruit putatively scene-selective regions of the human brain (the parahippocampal/lingual region, the retrosplenial complex, and the transverse occipital sulcus/occipital place area). Moreover, we find that the voxel-wise neural patterns arising from these associations are significantly correlated with the neural patterns arising from everyday scenes providing critical evidence whether the same encoding principals underlie both types of processing. These neuroimaging results provide evidence for the hypothesis that the neural representation of scenes is better understood within the broader theoretical framework of associative processing. In addition, the results demonstrate a division of labor that arises across scene-selective regions when processing associations and scenes providing better understanding of the functional roles of each region within the cortical network that mediates scene processing.     

Investigating the Neural Correlates of a Streaming Percept in an Informational-Masking Paradigm

Humans routinely segregate a complex acoustic scene into different auditory streams, through the extraction of bottom-up perceptual cues and the use of top-down selective attention. To determine the neural mechanisms underlying this process, neural responses obtained through magnetoencephalography (MEG) were correlated with behavioral performance in the context of an informational masking paradigm. In half the trials, subjects were asked to detect frequency deviants in a target stream, consisting of a rhythmic tone sequence, embedded in a separate masker stream composed of a random cloud of tones. In the other half of the trials, subjects were exposed to identical stimuli but asked to perform a different task—to detect tone-length changes in the random cloud of tones. In order to verify that the normalized neural response to the target sequence served as an indicator of streaming, we correlated neural responses with behavioral performance under a variety of stimulus parameters (target tone rate, target tone frequency, and the “protection zone”, that is, the spectral area with no tones around the target frequency) and attentional states (changing task objective while maintaining the same stimuli). In all conditions that facilitated target/masker streaming behaviorally, MEG normalized neural responses also changed in a manner consistent with the behavior. Thus, attending to the target stream caused a significant increase in power and phase coherence of the responses in recording channels correlated with an increase in the behavioral performance of the listeners. Normalized neural target responses also increased as the protection zone widened and as the frequency of the target tones increased. Finally, when the target sequence rate increased, the buildup of the normalized neural responses was significantly faster, mirroring the accelerated buildup of the streaming percepts. Our data thus support close links between the perceptual and neural consequences of the auditory stream segregation.     

Coding "what" and "when" in the Archer fish retina

Traditionally, the information content of the neural response is quantified using statistics of the responses relative to stimulus onset time with the assumption that the brain uses onset time to infer stimulus identity. However, stimulus onset time must also be estimated by the brain, making the utility of such an approach questionable. How can stimulus onset be estimated from the neural responses with sufficient accuracy to ensure reliable stimulus identification? We address this question using the framework of colour coding by the archer fish retinal ganglion cell. We found that stimulus identity, “what”, can be estimated from the responses of best single cells with an accuracy comparable to that of the animal''s psychophysical estimation. However, to extract this information, an accurate estimation of stimulus onset is essential. We show that stimulus onset time, “when”, can be estimated using a linear-nonlinear readout mechanism that requires the response of a population of 100 cells. Thus, stimulus onset time can be estimated using a relatively simple readout. However, large nerve cell populations are required to achieve sufficient accuracy.

Authors Summary

In our interaction with the environment we are flooded with a stream of numerous objects and events. Our brain needs to understand the nature of these complex and rich stimuli in order to react. Research has shown ways in which a ‘what’ stimulus was presented can be encoded by the neural responses. However, to understand ‘what was the nature of the stimulus’ the brain needs to know ‘when’ the stimulus was presented. Here, we investigated how the onset of visual stimulus can be signalled by the retina to higher brain regions. We used archer fish as a framework to test the notion that the answer to the question of ‘when’ something has been presented lies within the larger cell population, whereas the answer to the question of ‘what’ has been presented may be found at the single-neuron level. The utility of the archer fish as model animal stems from its remarkable ability to shoot down insects settling on the foliage above the water level, and its ability to distinguish between artificial targets. Thus, the archer fish can provide the fish equivalent of a monkey or a human that can report psychophysical decisions.     

Within- and Cross-Modal Distance Information Disambiguate Visual Size-Change Perception

Perception is fundamentally underconstrained because different combinations of object properties can generate the same sensory information. To disambiguate sensory information into estimates of scene properties, our brains incorporate prior knowledge and additional “auxiliary” (i.e., not directly relevant to desired scene property) sensory information to constrain perceptual interpretations. For example, knowing the distance to an object helps in perceiving its size. The literature contains few demonstrations of the use of prior knowledge and auxiliary information in combined visual and haptic disambiguation and almost no examination of haptic disambiguation of vision beyond “bistable” stimuli. Previous studies have reported humans integrate multiple unambiguous sensations to perceive single, continuous object properties, like size or position. Here we test whether humans use visual and haptic information, individually and jointly, to disambiguate size from distance. We presented participants with a ball moving in depth with a changing diameter. Because no unambiguous distance information is available under monocular viewing, participants rely on prior assumptions about the ball''s distance to disambiguate their -size percept. Presenting auxiliary binocular and/or haptic distance information augments participants'' prior distance assumptions and improves their size judgment accuracy—though binocular cues were trusted more than haptic. Our results suggest both visual and haptic distance information disambiguate size perception, and we interpret these results in the context of probabilistic perceptual reasoning.     

Same or different? Abstract relational concept use in juvenile bamboo sharks and Malawi cichlids

Sorting objects and events into categories and concepts is an important cognitive prerequisite that spares an individual the learning of every object or situation encountered in its daily life. Accordingly, specific items are classified in general groups that allow fast responses to novel situations. The present study assessed whether bamboo sharks Chiloscyllium griseum and Malawi cichlids Pseudotropheus zebra can distinguish sets of stimuli (each stimulus consisting of two abstract, geometric objects) that meet two conceptual preconditions, i.e., (1) “sameness” versus “difference” and (2) a certain spatial arrangement of both objects. In two alternative forced choice experiments, individuals were first trained to choose two different, vertically arranged objects from two different but horizontally arranged ones. Pair discriminations were followed by extensive transfer test experiments. Transfer tests using stimuli consisting of (a) black and gray circles and (b) squares with novel geometric patterns provided conflicting information with respect to the learnt rule “choose two different, vertically arranged objects”, thereby investigating (1) the individuals’ ability to transfer previously gained knowledge to novel stimuli and (2) the abstract relational concept(s) or rule(s) applied to categorize these novel objects. Present results suggest that the level of processing and usage of both abstract concepts differed considerably between bamboo sharks and Malawi cichlids. Bamboo sharks seemed to combine both concepts—although not with equal but hierarchical prominence—pointing to advanced cognitive capabilities. Conversely, Malawi cichlids had difficulties in discriminating between symbols and failed to apply the acquired training knowledge on new sets of geometric and, in particular, gray-level transfer stimuli.     

The color phi phenomenon: Not so special,after all?

We show how anomalous time reversal of stimuli and their associated responses can exist in very small connectionist models. These networks are built from dynamical toy model neurons which adhere to a minimal set of biologically plausible properties. The appearance of a “ghost” response, temporally and spatially located in between responses caused by actual stimuli, as in the phi phenomenon, is demonstrated in a similar small network, where it is caused by priming and long-distance feedforward paths. We then demonstrate that the color phi phenomenon can be present in an echo state network, a recurrent neural network, without explicitly training for the presence of the effect, such that it emerges as an artifact of the dynamical processing. Our results suggest that the color phi phenomenon might simply be a feature of the inherent dynamical and nonlinear sensory processing in the brain and in and of itself is not related to consciousness.     

Hippocampus Is Place of Interaction between Unconscious and Conscious Memories

Recent evidence suggests that humans can form and later retrieve new semantic relations unconsciously by way of hippocampus—the key structure also recruited for conscious relational (episodic) memory. If the hippocampus subserves both conscious and unconscious relational encoding/retrieval, one would expect the hippocampus to be place of unconscious-conscious interactions during memory retrieval. We tested this hypothesis in an fMRI experiment probing the interaction between the unconscious and conscious retrieval of face-associated information. For the establishment of unconscious relational memories, we presented subliminal (masked) combinations of unfamiliar faces and written occupations (“actor” or “politician”). At test, we presented the former subliminal faces, but now supraliminally, as cues for the reactivation of the unconsciously associated occupations. We hypothesized that unconscious reactivation of the associated occupation—actor or politician—would facilitate or inhibit the subsequent conscious retrieval of a celebrity’s occupation, which was also actor or politician. Depending on whether the reactivated unconscious occupation was congruent or incongruent to the celebrity’s occupation, we expected either quicker or delayed conscious retrieval process. Conscious retrieval was quicker in the congruent relative to a neutral baseline condition but not delayed in the incongruent condition. fMRI data collected during subliminal face-occupation encoding confirmed previous evidence that the hippocampus was interacting with neocortical storage sites of semantic knowledge to support relational encoding. fMRI data collected at test revealed that the facilitated conscious retrieval was paralleled by deactivations in the hippocampus and neocortical storage sites of semantic knowledge. We assume that the unconscious reactivation has pre-activated overlapping relational representations in the hippocampus reducing the neural effort for conscious retrieval. This finding supports the notion of synergistic interactions between conscious and unconscious relational memories in a common, cohesive hippocampal-neocortical memory space.     

The thymic cortical epithelium determines the TCR repertoire of IL-17-producing γδT cells

The thymus provides a specialized microenvironment in which distinct subsets of thymic epithelial cells (TECs) support T-cell development. Here, we describe the significance of cortical TECs (cTECs) in T-cell development, using a newly established mouse model of cTEC deficiency. The deficiency of mature cTECs caused a massive loss of thymic cellularity and impaired the development of αβT cells and invariant natural killer T cells. Unexpectedly, the differentiation of certain γδT-cell subpopulations—interleukin-17-producing Vγ4 and Vγ6 cells—was strongly dysregulated, resulting in the perturbation of γδT-mediated inflammatory responses in peripheral tissues. These findings show that cTECs contribute to the shaping of the TCR repertoire, not only of “conventional” αβT cells but also of inflammatory “innate” γδT cells.