Unsupervised learning of mid-level visual representations.

Recently, a confluence between trends in neuroscience and machine learning has brought a renewed focus on unsupervised learning, where sensory processing systems learn to exploit the statistical structure of their inputs in the absence of explicit training targets or rewards. Sophisticated experimental approaches have enabled the investigation of the influence of sensory experience on neural self-organization and its synaptic bases. Meanwhile, novel algorithms for unsupervised and self-supervised learning have become increasingly popular both as inspiration for theories of the brain, particularly for the function of intermediate visual cortical areas, and as building blocks of real-world learning machines. Here we review some of these recent developments, placing them in historical context and highlighting some research lines that promise exciting breakthroughs in the near future.

Truly pattern: Nonlinear integration of motion signals is required to account for the responses of pattern cells in rat visual cortex.

A key feature of advanced motion processing in the primate dorsal stream is the existence of pattern cells-specialized cortical neurons that integrate local motion signals into pattern-invariant representations of global direction. Pattern cells have also been reported in rodent visual cortex, but it is unknown whether the tuning of these neurons results from truly integrative, nonlinear mechanisms or trivially arises from linear receptive fields (RFs) with a peculiar geometry. Here, we show that pattern cells in rat primary (V1) and lateromedial (LM) visual cortex process motion direction in a way that cannot be explained by the linear spatiotemporal structure of their RFs. Instead, their tuning properties are consistent with and well explained by those of units in a state-of-the-art neural network model of the dorsal stream. This suggests that similar cortical processes underlay motion representation in primates and rodents. The latter could thus serve as powerful model systems to unravel the underlying circuit-level mechanisms.

Rat sensitivity to multipoint statistics is predicted by efficient coding of natural scenes.

Efficient processing of sensory data requires adapting the neuronal encoding strategy to the statistics of natural stimuli. Previously, in Hermundstad et al., 2014, we showed that local multipoint correlation patterns that are most variable in natural images are also the most perceptually salient for human observers, in a way that is compatible with the efficient coding principle. Understanding the neuronal mechanisms underlying such adaptation to image statistics will require performing invasive experiments that are impossible in humans. Therefore, it is important to understand whether a similar phenomenon can be detected in animal species that allow for powerful experimental manipulations, such as rodents. Here we selected four image statistics (from single- to four-point correlations) and trained four groups of rats to discriminate between white noise patterns and binary textures containing variable intensity levels of one of such statistics. We interpreted the resulting psychometric data with an ideal observer model, finding a sharp decrease in sensitivity from two- to four-point correlations and a further decrease from four- to three-point. This ranking fully reproduces the trend we previously observed in humans, thus extending a direct demonstration of efficient coding to a species where neuronal and developmental processes can be interrogated and causally manipulated.

Rats spontaneously perceive global motion direction of drifting plaids.

Computing global motion direction of extended visual objects is a hallmark of primate high-level vision. Although neurons selective for global motion have also been found in mouse visual cortex, it remains unknown whether rodents can combine multiple motion signals into global, integrated percepts. To address this question, we trained two groups of rats to discriminate either gratings (G group) or plaids (i.e., superpositions of gratings with different orientations; P group) drifting horizontally along opposite directions. After the animals learned the task, we applied a visual priming paradigm, where presentation of the target stimulus was preceded by the brief presentation of either a grating or a plaid. The extent to which rat responses to the targets were biased by such prime stimuli provided a measure of the spontaneous, perceived similarity between primes and targets. We found that gratings and plaids, when used as primes, were equally effective at biasing the perception of plaid direction for the rats of the P group. Conversely, for the G group, only the gratings acted as effective prime stimuli, while the plaids failed to alter the perception of grating direction. To interpret these observations, we simulated a decision neuron reading out the representations of gratings and plaids, as conveyed by populations of either component or pattern cells (i.e., local or global motion detectors). We concluded that the findings for the P group are highly consistent with the existence of a population of pattern cells, playing a functional role similar to that demonstrated in primates. We also explored different scenarios that could explain the failure of the plaid stimuli to elicit a sizable priming magnitude for the G group. These simulations yielded testable predictions about the properties of motion representations in rodent visual cortex at the single-cell and circuitry level, thus paving the way to future neurophysiology experiments.

Temporal stability of stimulus representation increases along rodent visual cortical hierarchies.

Cortical representations of brief, static stimuli become more invariant to identity-preserving transformations along the ventral stream. Likewise, increased invariance along the visual hierarchy should imply greater temporal persistence of temporally structured dynamic stimuli, possibly complemented by temporal broadening of neuronal receptive fields. However, such stimuli could engage adaptive and predictive processes, whose impact on neural coding dynamics is unknown. By probing the rat analog of the ventral stream with movies, we uncovered a hierarchy of temporal scales, with deeper areas encoding visual information more persistently. Furthermore, the impact of intrinsic dynamics on the stability of stimulus representations grew gradually along the hierarchy. A database of recordings from mouse showed similar trends, additionally revealing dependencies on the behavioral state. Overall, these findings show that visual representations become progressively more stable along rodent visual processing hierarchies, with an important contribution provided by intrinsic processing.

A machine learning framework to optimize optic nerve electrical stimulation for vision restoration.

Optic nerve electrical stimulation is a promising technique to restore vision in blind subjects. Machine learning methods can be used to select effective stimulation protocols, but they require a model of the stimulated system to generate enough training data. Here, we use a convolutional neural network (CNN) as a model of the ventral visual stream. A genetic algorithm drives the activation of the units in a layer of the CNN representing a cortical region toward a desired pattern, by refining the activation imposed at a layer representing the optic nerve. To simulate the pattern of activation elicited by the sites of an electrode array, a simple point-source model was introduced and its optimization process was investigated for static and dynamic scenes. Psychophysical data confirm that our stimulation evolution framework produces results compatible with natural vision. Machine learning approaches could become a very powerful tool to optimize and personalize neuroprosthetic systems.

A general-purpose mechanism of visual feature association in visual word identification and beyond.

As writing systems are a relatively novel invention (slightly over 5 kya),1 they could not have influenced the evolution of our species. Instead, reading might recycle evolutionary older mechanisms that originally supported other tasks2,3 and preceded the emergence of written language. Accordingly, it has been shown that baboons and pigeons can be trained to distinguish words from nonwords based on orthographic regularities in letter co-occurrence.4,5 This suggests that part of what is usually considered reading-specific processing could be performed by domain-general visual mechanisms. Here, we tested this hypothesis in humans: if the reading system relies on domain-general visual mechanisms, some of the effects that are often found with orthographic material should also be observable with non-orthographic visual stimuli. We performed three experiments using the same exact design but with visual stimuli that progressively departed from orthographic material. Subjects were passively familiarized with a set of composite visual items and tested in an oddball paradigm for their ability to detect novel stimuli. Participants showed robust sensitivity to the co-occurrence of features ("bigram" coding) with strings of letter-like symbols but also with made-up 3D objects and sinusoidal gratings. This suggests that the processing mechanisms involved in the visual recognition of novel words also support the recognition of other novel visual objects. These mechanisms would allow the visual system to capture statistical regularities in the visual environment.6-9 We hope that this work will inspire models of reading that, although addressing its unique aspects, place it within the broader context of vision.

A template-matching algorithm for laminar identification of cortical recording sites from evoked response potentials.

In recent years, the advent of the so-called silicon probes has made it possible to homogeneously sample spikes and local field potentials (LFPs) from a regular grid of cortical recording sites. In principle, this allows inferring the laminar location of the sites based on the spatiotemporal pattern of LFPs recorded along the probe, as in the well-known current source-density (CSD) analysis. This approach, however, has several limitations, since it relies on visual identification of landmark features (i.e., current sinks and sources) by human operators - features that can be absent from the CSD pattern if the probe does not span the whole cortical thickness, thus making manual labelling harder. Furthermore, as any manual annotation procedure, the typical CSD-based workflow for laminar identification of recording sites is affected by subjective judgment undermining the consistency and reproducibility of results. To overcome these limitations, we developed an alternative approach, based on finding the optimal match between the LFPs recorded along a probe in a given experiment and a template LFP profile that was computed using 18 recording sessions, in which the depth of the recording sites had been recovered through histology. We show that this method can achieve an accuracy of 79 µm in recovering the cortical depth of recording sites and a 76% accuracy in inferring their laminar location. As such, our approach provides an alternative to CSD that, being fully automated, is less prone to the idiosyncrasies of subjective judgment and works reliably also for recordings spanning a limited cortical stretch.

Unsupervised experience with temporal continuity of the visual environment is causally involved in the development of V1 complex cells.

Unsupervised adaptation to the spatiotemporal statistics of visual experience is a key computational principle that has long been assumed to govern postnatal development of visual cortical tuning, including orientation selectivity of simple cells and position tolerance of complex cells in primary visual cortex (V1). Yet, causal empirical evidence supporting this hypothesis is scant. Here, we show that degrading the temporal continuity of visual experience during early postnatal life leads to a sizable reduction of the number of complex cells and to an impairment of their functional properties while fully sparing the development of simple cells. This causally implicates adaptation to the temporal structure of the visual input in the development of transformation tolerance but not of shape tuning, thus tightly constraining computational models of unsupervised cortical learning.

A passive, camera-based head-tracking system for real-time, three-dimensional estimation of head position and orientation in rodents.

Tracking head position and orientation in small mammals is crucial for many applications in the field of behavioral neurophysiology, from the study of spatial navigation to the investigation of active sensing and perceptual representations. Many approaches to head tracking exist, but most of them only estimate the 2D coordinates of the head over the plane where the animal navigates. Full reconstruction of the pose of the head in 3D is much more more challenging and has been achieved only in handful of studies, which employed headsets made of multiple LEDs or inertial units. However, these assemblies are rather bulky and need to be powered to operate, which prevents their application in wireless experiments and in the small enclosures often used in perceptual studies. Here we propose an alternative approach, based on passively imaging a lightweight, compact, 3D structure, painted with a pattern of black dots over a white background. By applying a cascade of feature extraction algorithms that progressively refine the detection of the dots and reconstruct their geometry, we developed a tracking method that is highly precise and accurate, as assessed through a battery of validation measurements. We show that this method can be used to study how a rat samples sensory stimuli during a perceptual discrimination task and how a hippocampal place cell represents head position over extremely small spatial scales. Given its minimal encumbrance and wireless nature, our method could be ideal for high-throughput applications, where tens of animals need to be simultaneously and continuously tracked.NEW & NOTEWORTHY Head tracking is crucial in many behavioral neurophysiology studies. Yet reconstruction of the head's pose in 3D is challenging and typically requires implanting bulky, electrically powered headsets that prevent wireless experiments and are hard to employ in operant boxes. Here we propose an alternative approach, based on passively imaging a compact, 3D dot pattern that, once implanted over the head of a rodent, allows estimating the pose of its head with high precision and accuracy.

Nonlinear Processing of Shape Information in Rat Lateral Extrastriate Cortex.

In rodents, the progression of extrastriate areas located laterally to primary visual cortex (V1) has been assigned to a putative object-processing pathway (homologous to the primate ventral stream), based on anatomical considerations. Recently, we found functional support for such attribution (Tafazoli et al., 2017), by showing that this cortical progression is specialized for coding object identity despite view changes, the hallmark property of a ventral-like pathway. Here, we sought to clarify what computations are at the base of such specialization. To this aim, we performed multielectrode recordings from V1 and laterolateral area LL (at the apex of the putative ventral-like hierarchy) of male adult rats, during the presentation of drifting gratings and noise movies. We found that the extent to which neuronal responses were entrained to the phase of the gratings sharply dropped from V1 to LL, along with the quality of the receptive fields inferred through reverse correlation. Concomitantly, the tendency of neurons to respond to different oriented gratings increased, whereas the sharpness of orientation tuning declined. Critically, these trends are consistent with the nonlinear summation of visual inputs that is expected to take place along the ventral stream, according to the predictions of hierarchical models of ventral computations and a meta-analysis of the monkey literature. This suggests an intriguing homology between the mechanisms responsible for building up shape selectivity and transformation tolerance in the visual cortex of primates and rodents, reasserting the potential of the latter as models to investigate ventral stream functions at the circuitry level.SIGNIFICANCE STATEMENT Despite the growing popularity of rodents as models of visual functions, it remains unclear whether their visual cortex contains specialized modules for processing shape information. To addresses this question, we compared how neuronal tuning evolves from rat primary visual cortex (V1) to a downstream visual cortical region (area LL) that previous work has implicated in shape processing. In our experiments, LL neurons displayed a stronger tendency to respond to drifting gratings with different orientations while maintaining a sustained response across the whole duration of the drift cycle. These trends match the increased complexity of pattern selectivity and the augmented tolerance to stimulus translation found in monkey visual temporal cortex, thus revealing a homology between shape processing in rodents and primates.

Accuracy of Rats in Discriminating Visual Objects Is Explained by the Complexity of Their Perceptual Strategy.

Despite their growing popularity as models of visual functions, it remains unclear whether rodents are capable of deploying advanced shape-processing strategies when engaged in visual object recognition. In rats, for instance, pattern vision has been reported to range from mere detection of overall object luminance to view-invariant processing of discriminative shape features. Here we sought to clarify how refined object vision is in rodents, and how variable the complexity of their visual processing strategy is across individuals. To this aim, we measured how well rats could discriminate a reference object from 11 distractors, which spanned a spectrum of image-level similarity to the reference. We also presented the animals with random variations of the reference, and processed their responses to these stimuli to derive subject-specific models of rat perceptual choices. Our models successfully captured the highly variable discrimination performance observed across subjects and object conditions. In particular, they revealed that the animals that succeeded with the most challenging distractors were those that integrated the wider variety of discriminative features into their perceptual strategies. Critically, these strategies were largely preserved when the rats were required to discriminate outlined and scaled versions of the stimuli, thus showing that rat object vision can be characterized as a transformation-tolerant, feature-based filtering process. Overall, these findings indicate that rats are capable of advanced processing of shape information, and point to the rodents as powerful models for investigating the neuronal underpinnings of visual object recognition and other high-level visual functions.

Supralinear and Supramodal Integration of Visual and Tactile Signals in Rats: Psychophysics and Neuronal Mechanisms.

To better understand how object recognition can be triggered independently of the sensory channel through which information is acquired, we devised a task in which rats judged the orientation of a raised, black and white grating. They learned to recognize two categories of orientation: 0° ± 45° ("horizontal") and 90° ± 45° ("vertical"). Each trial required a visual (V), a tactile (T), or a visual-tactile (VT) discrimination; VT performance was better than that predicted by optimal linear combination of V and T signals, indicating synergy between sensory channels. We examined posterior parietal cortex (PPC) and uncovered key neuronal correlates of the behavioral findings: PPC carried both graded information about object orientation and categorical information about the rat's upcoming choice; single neurons exhibited identical responses under the three modality conditions. Finally, a linear classifier of neuronal population firing replicated the behavioral findings. Taken together, these findings suggest that PPC is involved in the supramodal processing of shape.

Emergence of transformation-tolerant representations of visual objects in rat lateral extrastriate cortex.

Rodents are emerging as increasingly popular models of visual functions. Yet, evidence that rodent visual cortex is capable of advanced visual processing, such as object recognition, is limited. Here we investigate how neurons located along the progression of extrastriate areas that, in the rat brain, run laterally to primary visual cortex, encode object information. We found a progressive functional specialization of neural responses along these areas, with: (1) a sharp reduction of the amount of low-level, energy-related visual information encoded by neuronal firing; and (2) a substantial increase in the ability of both single neurons and neuronal populations to support discrimination of visual objects under identity-preserving transformations (e.g., position and size changes). These findings strongly argue for the existence of a rat object-processing pathway, and point to the rodents as promising models to dissect the neuronal circuitry underlying transformation-tolerant recognition of visual objects.

3D meshes of carbon nanotubes guide functional reconnection of segregated spinal explants.

In modern neuroscience, significant progress in developing structural scaffolds integrated with the brain is provided by the increasing use of nanomaterials. We show that a multiwalled carbon nanotube self-standing framework, consisting of a three-dimensional (3D) mesh of interconnected, conductive, pure carbon nanotubes, can guide the formation of neural webs in vitro where the spontaneous regrowth of neurite bundles is molded into a dense random net. This morphology of the fiber regrowth shaped by the 3D structure supports the successful reconnection of segregated spinal cord segments. We further observed in vivo the adaptability of these 3D devices in a healthy physiological environment. Our study shows that 3D artificial scaffolds may drive local rewiring in vitro and hold great potential for the development of future in vivo interfaces.

Object similarity affects the perceptual strategy underlying invariant visual object recognition in rats.

In recent years, a number of studies have explored the possible use of rats as models of high-level visual functions. One central question at the root of such an investigation is to understand whether rat object vision relies on the processing of visual shape features or, rather, on lower-order image properties (e.g., overall brightness). In a recent study, we have shown that rats are capable of extracting multiple features of an object that are diagnostic of its identity, at least when those features are, structure-wise, distinct enough to be parsed by the rat visual system. In the present study, we have assessed the impact of object structure on rat perceptual strategy. We trained rats to discriminate between two structurally similar objects, and compared their recognition strategies with those reported in our previous study. We found that, under conditions of lower stimulus discriminability, rat visual discrimination strategy becomes more view-dependent and subject-dependent. Rats were still able to recognize the target objects, in a way that was largely tolerant (i.e., invariant) to object transformation; however, the larger structural and pixel-wise similarity affected the way objects were processed. Compared to the findings of our previous study, the patterns of diagnostic features were: (i) smaller and more scattered; (ii) only partially preserved across object views; and (iii) only partially reproducible across rats. On the other hand, rats were still found to adopt a multi-featural processing strategy and to make use of part of the optimal discriminatory information afforded by the two objects. Our findings suggest that, as in humans, rat invariant recognition can flexibly rely on either view-invariant representations of distinctive object features or view-specific object representations, acquired through learning.

Invariant visual object recognition and shape processing in rats.

Invariant visual object recognition is the ability to recognize visual objects despite the vastly different images that each object can project onto the retina during natural vision, depending on its position and size within the visual field, its orientation relative to the viewer, etc. Achieving invariant recognition represents such a formidable computational challenge that is often assumed to be a unique hallmark of primate vision. Historically, this has limited the invasive investigation of its neuronal underpinnings to monkey studies, in spite of the narrow range of experimental approaches that these animal models allow. Meanwhile, rodents have been largely neglected as models of object vision, because of the widespread belief that they are incapable of advanced visual processing. However, the powerful array of experimental tools that have been developed to dissect neuronal circuits in rodents has made these species very attractive to vision scientists too, promoting a new tide of studies that have started to systematically explore visual functions in rats and mice. Rats, in particular, have been the subjects of several behavioral studies, aimed at assessing how advanced object recognition and shape processing is in this species. Here, I review these recent investigations, as well as earlier studies of rat pattern vision, to provide an historical overview and a critical summary of the status of the knowledge about rat object vision. The picture emerging from this survey is very encouraging with regard to the possibility of using rats as complementary models to monkeys in the study of higher-level vision.

Shape similarity, better than semantic membership, accounts for the structure of visual object representations in a population of monkey inferotemporal neurons.

The anterior inferotemporal cortex (IT) is the highest stage along the hierarchy of visual areas that, in primates, processes visual objects. Although several lines of evidence suggest that IT primarily represents visual shape information, some recent studies have argued that neuronal ensembles in IT code the semantic membership of visual objects (i.e., represent conceptual classes such as animate and inanimate objects). In this study, we investigated to what extent semantic, rather than purely visual information, is represented in IT by performing a multivariate analysis of IT responses to a set of visual objects. By relying on a variety of machine-learning approaches (including a cutting-edge clustering algorithm that has been recently developed in the domain of statistical physics), we found that, in most instances, IT representation of visual objects is accounted for by their similarity at the level of shape or, more surprisingly, low-level visual properties. Only in a few cases we observed IT representations of semantic classes that were not explainable by the visual similarity of their members. Overall, these findings reassert the primary function of IT as a conveyor of explicit visual shape information, and reveal that low-level visual properties are represented in IT to a greater extent than previously appreciated. In addition, our work demonstrates how combining a variety of state-of-the-art multivariate approaches, and carefully estimating the contribution of shape similarity to the representation of object categories, can substantially advance our understanding of neuronal coding of visual objects in cortex.