Distributed representation of visual objects by single neurons in the human brain

Encoding of Race Categories by Single Neurons in the Human Brain

Previous research has suggested that race-specific features are automatically processed during face perception, often with out-group faces treated categorically. Functional imaging has illuminated the hemodynamic correlates of this process, with fewer studies examining single-neuron responses. In the present experiment, epilepsy patients undergoing microwire recordings in preparation for surgical treatment were shown realistic computer-generated human faces, which they classified according to the emotional expression shown. Racial categories of the stimulus faces varied independently of the emotion shown, being irrelevant to the patients' primary task. Nevertheless, we observed race-driven changes in neural firing rates in the amygdala, anterior cingulate cortex, and hippocampus. These responses were broadly distributed, with the firing rates of 28% of recorded neurons in the amygdala and 45% in the anterior cingulate cortex predicting one or more racial categories. Nearly equal proportions of neurons responded to White and Black faces (24% vs. 22% in the amygdala and 26% vs. 28% in the anterior cingulate cortex). A smaller fraction (12%) of race-responsive neurons in the hippocampus predicted only White faces. Our results imply a distributed representation of race in brain areas involved in affective judgments, decision making, and memory. They also support the hypothesis that race-specific cues are perceptually coded even when those cues are task-irrelevant.

Toward Reflective Spiking Neural Networks Exploiting Memristive Devices

The design of modern convolutional artificial neural networks (ANNs) composed of formal neurons copies the architecture of the visual cortex. Signals proceed through a hierarchy, where receptive fields become increasingly more complex and coding sparse. Nowadays, ANNs outperform humans in controlled pattern recognition tasks yet remain far behind in cognition. In part, it happens due to limited knowledge about the higher echelons of the brain hierarchy, where neurons actively generate predictions about what will happen next, i.e., the information processing jumps from reflex to reflection. In this study, we forecast that spiking neural networks (SNNs) can achieve the next qualitative leap. Reflective SNNs may take advantage of their intrinsic dynamics and mimic complex, not reflex-based, brain actions. They also enable a significant reduction in energy consumption. However, the training of SNNs is a challenging problem, strongly limiting their deployment. We then briefly overview new insights provided by the concept of a high-dimensional brain, which has been put forward to explain the potential power of single neurons in higher brain stations and deep SNN layers. Finally, we discuss the prospect of implementing neural networks in memristive systems. Such systems can densely pack on a chip 2D or 3D arrays of plastic synaptic contacts directly processing analog information. Thus, memristive devices are a good candidate for implementing in-memory and in-sensor computing. Then, memristive SNNs can diverge from the development of ANNs and build their niche, cognitive, or reflective computations.

What kind of network is the brain?

The different areas of the cerebral cortex are linked by a network of white matter, comprising the myelinated axons of pyramidal cells. Is this network a neural net, in the sense that representations of the world are embodied in the structure of the net, its pattern of nodes, and connections? Or is it a communications network, where the same physical substrate carries different information from moment to moment? This question is part of the larger question of whether the brain is better modeled by connectionism or by symbolic artificial intelligence (AI), but we review it in the specific context of the psychophysics of stimulus comparison and the format and protocol of information transmission over the long-range tracts of the brain.

Estimates of distributed coding of visual objects by single neurons in the human brain depend on which spike sorting technique is used

OBJECTIVE

To determine whether the estimated fraction and degree of distribution of visual responsive neurons in human intracranial microwire recordings depends upon the spike sorting method used.

APPROACH

A large dataset of human intracranial microwire recordings from four brain areas was sorted into single unit activity (SUA) and multiunit activity (MUA) using 4 spike sorting methods previously applied to this type of recording. The responses were examined for visual responses to 33 objects which were presented.

MAIN RESULTS

The 4 spike sorting techniques examined here yielded fractions of responsive SUA varying from 8% in the left anterior cingulate cortex to 27% in the right amgdala. The fraction of responsive SUA and MUA depended on the type of spike sorting being used as well as brain area and side being recorded from. Agreement between spike sorting techniques was low (0.04-0.16 on the 0-1 AMI_all scale).

SIGNIFICANCE

Prior estimates of the fraction of single neurons in the human medial temporal lobe coding semantic memory of visual objects have yielded fractions ranging from 0.04% by very strict response criteria to 47% by other criteria. A variety of explanations of these differences have been posited, including differences in the type of memory being tested, differences in visual stimuli, as well as technical differences such as spike sorting techniques. This study shows the dependence of the reported fraction of neurons encoding visual objects on the spike sorting technique employed and confirms a distributed representation of visual objects by single neurons in the human brain. Both the variation in the responsive fractions with spike sorting technique and low levels of agreement between techniques highlight the need for better understanding of the signals being extracted in human intracranial microwire recordings.

Representation of abstract semantic knowledge in populations of human single neurons in the medial temporal lobe

Sensory experience elicits complex activity patterns throughout the neocortex. Projections from the neocortex converge onto the medial temporal lobe (MTL), in which distributed neocortical firing patterns are distilled into sparse representations. The precise nature of these neuronal representations is still unknown. Here, we show that population activity patterns in the MTL are governed by high levels of semantic abstraction. We recorded human single-unit activity in the MTL (4,917 units, 25 patients) while subjects viewed 100 images grouped into 10 semantic categories of 10 exemplars each. High levels of semantic abstraction were indicated by representational similarity analyses (RSAs) of patterns elicited by individual stimuli. Moreover, pattern classifiers trained to decode semantic categories generalised successfully to unseen exemplars, and classifiers trained to decode exemplar identity more often confused exemplars of the same versus different categories. Semantic abstraction and generalisation may thus be key to efficiently distill the essence of an experience into sparse representations in the human MTL. Although semantic abstraction is efficient and may facilitate generalisation of knowledge to novel situations, it comes at the cost of a loss of detail and may be central to the generation of false memories.

Single-neuron representations of stimuli in the human medial temporal lobe at the population level are governed by highly abstract semantic principles, but the attendant efficiency and potential for generalization comes at the cost of confusion between related stimuli.

What is the neuronal code for sensory experience in the human medial temporal lobe (MTL)? Single-cell electrophysiology in the awake human brain during chronic, invasive epilepsy monitoring has previously revealed the existence of so-called concept cells. These cells have been found to increase their firing rate in response to, for example, the famous tennis player ‘Roger Federer’, whether his name is spoken by a computer voice or a picture of him is presented on a computer screen. These neurons thus seem to encode the semantic content of a stimulus, regardless of the sensory modality through which it is delivered. Previous work has predominantly focused on individual neurons that were selected based on their strong response to a particular stimulus using rather conservative statistical criteria. Those studies stressed that concept cells encode a single, concrete concept in an all-or-nothing fashion. Here, we analysed the neuronal code on the level of the entire population of neurons without any preselection. We conducted representational similarity analyses (RSAs) and pattern classification analyses of firing patterns evoked by visual stimuli (for example, a picture of an apple) that could be grouped into semantic categories on multiple levels of abstraction (‘fruit’, ‘food’, ‘natural things’). We found that neuronal activation patterns contain information on higher levels of categorical abstraction rather than just the level of individual exemplars. On the one hand, the neuronal code in the human MTL thus seems well suited to generalise semantic knowledge to new situations; on the other hand, it could also be responsible for the generation of false memories.

The unreasonable effectiveness of small neural ensembles in high-dimensional brain

Complexity is an indisputable, well-known, and broadly accepted feature of the brain. Despite the apparently obvious and widely-spread consensus on the brain complexity, sprouts of the single neuron revolution emerged in neuroscience in the 1970s. They brought many unexpected discoveries, including grandmother or concept cells and sparse coding of information in the brain. In machine learning for a long time, the famous curse of dimensionality seemed to be an unsolvable problem. Nevertheless, the idea of the blessing of dimensionality becomes gradually more and more popular. Ensembles of non-interacting or weakly interacting simple units prove to be an effective tool for solving essentially multidimensional and apparently incomprehensible problems. This approach is especially useful for one-shot (non-iterative) correction of errors in large legacy artificial intelligence systems and when the complete re-training is impossible or too expensive. These simplicity revolutions in the era of complexity have deep fundamental reasons grounded in geometry of multidimensional data spaces. To explore and understand these reasons we revisit the background ideas of statistical physics. In the course of the 20th century they were developed into the concentration of measure theory. The Gibbs equivalence of ensembles with further generalizations shows that the data in high-dimensional spaces are concentrated near shells of smaller dimension. New stochastic separation theorems reveal the fine structure of the data clouds. We review and analyse biological, physical, and mathematical problems at the core of the fundamental question: how can high-dimensional brain organise reliable and fast learning in high-dimensional world of data by simple tools? To meet this challenge, we outline and setup a framework based on statistical physics of data. Two critical applications are reviewed to exemplify the approach: one-shot correction of errors in intellectual systems and emergence of static and associative memories in ensembles of single neurons. Error correctors should be simple; not damage the existing skills of the system; allow fast non-iterative learning and correction of new mistakes without destroying the previous fixes. All these demands can be satisfied by new tools based on the concentration of measure phenomena and stochastic separation theory. We show how a simple enough functional neuronal model is capable of explaining: i) the extreme selectivity of single neurons to the information content of high-dimensional data, ii) simultaneous separation of several uncorrelated informational items from a large set of stimuli, and iii) dynamic learning of new items by associating them with already "known" ones. These results constitute a basis for organisation of complex memories in ensembles of single neurons.

Precision and False Perceptual Inference

Accurate perceptual inference fundamentally depends upon accurate beliefs about the reliability of sensory data. In this paper, we describe a Bayes optimal and biologically plausible scheme that refines these beliefs through a gradient descent on variational free energy. To illustrate this, we simulate belief updating during visual foraging and show that changes in estimated sensory precision (i.e., confidence in visual data) are highly sensitive to prior beliefs about the contents of a visual scene. In brief, confident prior beliefs induce an increase in estimated precision when consistent with sensory evidence, but a decrease when they conflict. Prior beliefs held with low confidence are rapidly updated to posterior beliefs, determined by sensory data. These induce much smaller changes in beliefs about sensory precision. We argue that pathologies of scene construction may be due to abnormal priors, and show that these can induce a reduction in estimated sensory precision. Having previously associated this precision with cholinergic signaling, we note that several neurodegenerative conditions are associated with visual disturbances and cholinergic deficits; notably, the synucleinopathies. On relating the message passing in our model to the functional anatomy of the ventral visual stream, we find that simulated neuronal loss in temporal lobe regions induces confident, inaccurate, empirical prior beliefs at lower levels in the visual hierarchy. This provides a plausible, if speculative, computational mechanism for the loss of cholinergic signaling and the visual disturbances associated with temporal lobe Lewy body pathology. This may be seen as an illustration of the sorts of hypotheses that may be expressed within this computational framework.

Computational Neuropsychology and Bayesian Inference

Computational theories of brain function have become very influential in neuroscience. They have facilitated the growth of formal approaches to disease, particularly in psychiatric research. In this paper, we provide a narrative review of the body of computational research addressing neuropsychological syndromes, and focus on those that employ Bayesian frameworks. Bayesian approaches to understanding brain function formulate perception and action as inferential processes. These inferences combine ‘prior’ beliefs with a generative (predictive) model to explain the causes of sensations. Under this view, neuropsychological deficits can be thought of as false inferences that arise due to aberrant prior beliefs (that are poor fits to the real world). This draws upon the notion of a Bayes optimal pathology – optimal inference with suboptimal priors – and provides a means for computational phenotyping. In principle, any given neuropsychological disorder could be characterized by the set of prior beliefs that would make a patient’s behavior appear Bayes optimal. We start with an overview of some key theoretical constructs and use these to motivate a form of computational neuropsychology that relates anatomical structures in the brain to the computations they perform. Throughout, we draw upon computational accounts of neuropsychological syndromes. These are selected to emphasize the key features of a Bayesian approach, and the possible types of pathological prior that may be present. They range from visual neglect through hallucinations to autism. Through these illustrative examples, we review the use of Bayesian approaches to understand the link between biology and computation that is at the heart of neuropsychology.

Coding of episodic memory in the human hippocampus

Neurocomputational models have long posited that episodic memories in the human hippocampus are represented by sparse, stimulus-specific neural codes. A concomitant proposal is that when sparse-distributed neural assemblies become active, they suppress the activity of competing neurons (neural sharpening). We investigated episodic memory coding in the hippocampus and amygdala by measuring single-neuron responses from 20 epilepsy patients (12 female) undergoing intracranial monitoring while they completed a continuous recognition memory task. In the left hippocampus, the distribution of single-neuron activity indicated that only a small fraction of neurons exhibited strong responding to a given repeated word and that each repeated word elicited strong responding in a different small fraction of neurons. This finding reflects sparse distributed coding. The remaining large fraction of neurons exhibited a concurrent reduction in firing rates relative to novel words. The observed pattern accords with longstanding predictions that have previously received scant support from single-cell recordings from human hippocampus.

Neuronal Assemblies Evidence Distributed Interactions within a Tactile Discrimination Task in Rats

Accumulating evidence suggests that neural interactions are distributed and relate to animal behavior, but many open questions remain. The neural assembly hypothesis, formulated by Hebb, states that synchronously active single neurons may transiently organize into functional neural circuits-neuronal assemblies (NAs)-and that would constitute the fundamental unit of information processing in the brain. However, the formation, vanishing, and temporal evolution of NAs are not fully understood. In particular, characterizing NAs in multiple brain regions over the course of behavioral tasks is relevant to assess the highly distributed nature of brain processing. In the context of NA characterization, active tactile discrimination tasks with rats are elucidative because they engage several cortical areas in the processing of information that are otherwise masked in passive or anesthetized scenarios. In this work, we investigate the dynamic formation of NAs within and among four different cortical regions in long-range fronto-parieto-occipital networks (primary somatosensory, primary visual, prefrontal, and posterior parietal cortices), simultaneously recorded from seven rats engaged in an active tactile discrimination task. Our results first confirm that task-related neuronal firing rate dynamics in all four regions is significantly modulated. Notably, a support vector machine decoder reveals that neural populations contain more information about the tactile stimulus than the majority of single neurons alone. Then, over the course of the task, we identify the emergence and vanishing of NAs whose participating neurons are shown to contain more information about animal behavior than randomly chosen neurons. Taken together, our results further support the role of multiple and distributed neurons as the functional unit of information processing in the brain (NA hypothesis) and their link to active animal behavior.

An online adaptive screening procedure for selective neuronal responses

BACKGROUND

A common problem in neurophysiology is to identify stimuli that elicit neuronal responses in a given brain region. Particularly in situations where electrode positions are fixed, this can be a time-consuming task that requires presentation of a large number of stimuli. Such a screening for response-eliciting stimuli is employed, e.g., as a standard procedure to identify 'concept cells' in the human medial temporal lobe.

NEW METHOD

Our new method evaluates neuronal responses to stimuli online during a screening session, which allows us to successively exclude stimuli that do not evoke a response. Using this method, we can screen a larger number of stimuli which in turn increases the chances of finding responsive neurons and renders time-consuming offline analysis unnecessary.

RESULTS

Our method enabled us to present 30% more stimuli in the same period of time with additional presentations of the most promising candidate stimuli. Our online method ran smoothly on a standard computer and network.

COMPARISON WITH AN EXISTING METHOD

To analyze how our online screening procedure performs in comparison to an established offline method, we used the Wave_Clus software package. We did not observe any major drawbacks in our method, but a much higher efficiency and analysis speed.

CONCLUSIONS

By transitioning from a traditional offline screening procedure to our new online method, we substantially increased the number of visual stimuli presented in a given time period. This allows to identify more response-eliciting stimuli, which forms the basis to better address a great number of questions in cognitive neuroscience.

Scene-selective coding by single neurons in the human parahippocampal cortex

Imaging, electrophysiological, and lesion studies have shown a relationship between the parahippocampal cortex (PHC) and the processing of spatial scenes. Our present knowledge of PHC, however, is restricted to the macroscopic properties and dynamics of bulk tissue; the behavior and selectivity of single parahippocampal neurons remains largely unknown. In this study, we analyzed responses from 630 parahippocampal neurons in 24 neurosurgical patients during visual stimulus presentation. We found a spatially clustered subpopulation of scene-selective units with an associated event-related field potential. These units form a population code that is more distributed for scenes than for other stimulus categories, and less sparse than elsewhere in the medial temporal lobe. Our electrophysiological findings provide insight into how individual units give rise to the population response observed with functional imaging in the parahippocampal place area.

Gender differences in human single neuron responses to male emotional faces

Well-documented differences in the psychology and behavior of men and women have spurred extensive exploration of gender's role within the brain, particularly regarding emotional processing. While neuroanatomical studies clearly show differences between the sexes, the functional effects of these differences are less understood. Neuroimaging studies have shown inconsistent locations and magnitudes of gender differences in brain hemodynamic responses to emotion. To better understand the neurophysiology of these gender differences, we analyzed recordings of single neuron activity in the human brain as subjects of both genders viewed emotional expressions. This study included recordings of single-neuron activity of 14 (6 male) epileptic patients in four brain areas: amygdala (236 neurons), hippocampus (n = 270), anterior cingulate cortex (n = 256), and ventromedial prefrontal cortex (n = 174). Neural activity was recorded while participants viewed a series of avatar male faces portraying positive, negative or neutral expressions. Significant gender differences were found in the left amygdala, where 23% (n = 15∕66) of neurons in men were significantly affected by facial emotion, vs. 8% (n = 6∕76) of neurons in women. A Fisher's exact test comparing the two ratios found a highly significant difference between the two (p < 0.01). These results show specific differences between genders at the single-neuron level in the human amygdala. These differences may reflect gender-based distinctions in evolved capacities for emotional processing and also demonstrate the importance of including subject gender as an independent factor in future studies of emotional processing by single neurons in the human amygdala.