Using deep neural networks to evaluate object vision tasks in rats.

In the last two decades rodents have been on the rise as a dominant model for visual neuroscience. This is particularly true for earlier levels of information processing, but a number of studies have suggested that also higher levels of processing such as invariant object recognition occur in rodents. Here we provide a quantitative and comprehensive assessment of this claim by comparing a wide range of rodent behavioral and neural data with convolutional deep neural networks. These networks have been shown to capture hallmark properties of information processing in primates through a succession of convolutional and fully connected layers. We find that performance on rodent object vision tasks can be captured using low to mid-level convolutional layers only, without any convincing evidence for the need of higher layers known to simulate complex object recognition in primates. Our approach also reveals surprising insights on assumptions made before, for example, that the best performing animals would be the ones using the most abstract representations-which we show to likely be incorrect. Our findings suggest a road ahead for further studies aiming at quantifying and establishing the richness of representations underlying information processing in animal models at large.

Intrinsic functional clustering of ventral premotor F5 in the macaque brain.

Neurophysiological and anatomical data suggest the existence of several functionally distinct regions in the lower arcuate sulcus and adjacent postarcuate convexity of the macaque monkey. Ventral premotor F5c lies on the postarcuate convexity and consists of a dorsal hand-related and ventral mouth-related field. The posterior bank of the lower arcuate contains two additional premotor F5 subfields at different anterior-posterior levels, F5a and F5p. Anterior to F5a, area 44 has been described as a dysgranular zone occupying the deepest part of the fundus of the inferior arcuate. Finally, area GrFO occupies the most rostral portion of the fundus and posterior bank of inferior arcuate and extends ventrally onto the frontal operculum. Recently, data-driven exploratory approaches using resting-state fMRI data have been suggested as a promising non-invasive method for examining the functional organization of the primate brain. Here, we examined to what extent partitioning schemes derived from data-driven clustering analysis of resting-state fMRI data correspond with the proposed organization of the fundus and posterior bank of the macaque arcuate sulcus, as suggested by invasive architectonical, connectional and functional investigations. Using a hierarchical clustering analysis, we could retrieve clusters corresponding to the dorsal and ventral portions of F5c on the postarcuate convexity, F5a and F5p at different antero-posterior locations on the posterior bank of the lower arcuate, area 44 in the fundus, as well as part of area GrFO in the most anterior portion of the fundus. Additionally, each of these clusters displayed distinct whole-brain functional connectivity, in line with previous anatomical tracer and seed-based functional connectivity investigations of F5/44 subdivisions. Overall, our data suggests that hierarchical clustering analysis of resting-state fMRI data can retrieve a fine-grained level of cortical organization that resembles detailed parcellation schemes derived from invasive functional and anatomical investigations.

Face Repetition Probability Does Not Affect Repetition Suppression in Macaque Inferotemporal Cortex.

Repetition suppression, which refers to reduced neural activity for repeated stimuli, is typically explained by bottom-up or local adaptation mechanisms. However, recent theories have emphasized the role of top-down processes, suggesting that this response reduction reflects the fulfillment of perceptual expectations. To support this, an influential human fMRI study showed that the magnitude of suppression is modulated by the probability of a repetition. No such repetition probability effect was found in macaque inferior temporal (IT) cortex for spiking activity despite the presence of repetition suppression. Contrary to the human fMRI studies that showed an effect of repetition probability, the macaque single-unit study used a large variety of unfamiliar stimuli and the monkeys were not required to attend the stimuli. Here, as in the human fMRI studies, we used faces as stimuli and made the monkeys attend to the stimulus content. We simultaneously recorded spiking activity and local field potentials (LFPs) in the middle lateral face patch (ML) of one monkey (male) and a face-responsive region of another (female). Although we observed significant repetition suppression of spiking activity and high gamma-band LFPs in both animals, there were no effects of repetition probability even when repetitions were task relevant and repetition probability affected behavioral decisions. In conclusion, despite the use of face stimuli and a stimulus-related task, no neural signature of repetition probability was present for faces in a face responsive patch of macaque IT. This further challenges a general perceptual expectation account of repetition suppression.SIGNIFICANCE STATEMENT Repetition suppression is a reduced brain activity for repeated stimuli commonly observed across species. In the predictive coding framework, such suppression is thought to reflect fulfilled perceptual expectations. Although this hypothesis is supported by several human fMRI studies reporting an effect of repetition probability on repetition suppression, this could not be replicated in single-cell recordings in monkey inferior temporal (IT) cortex. Subsequent studies narrowed down the conditions for the effect to requiring attention and being limited to particular stimulus categories such as faces. Here, we show that, even under these conditions, repetition suppression in monkey IT neurons is still unaffected by repetition probability, even in a task with a behavioral effect, challenging the perceptual expectation account of repetition suppression.

A behavioral face preference deficit in a monkey with an incomplete face patch system.

Primates are experts in face perception and naturally show a preference for faces under free-viewing conditions. The primate ventral stream is characterized by a network of face patches that selectively responds to faces, but it remains uncertain how important such parcellation is for face perception. Here we investigated free-viewing behavior in a female monkey who naturally lacks fMRI-defined posterior and middle lateral face patches. We presented a series of content-rich images of scenes that included faces or other objects to that monkey during a free-viewing task and tested a group of 10 control monkeys on the same task for comparison. We found that, compared to controls, the monkey with missing face patches showed a marked reduction of face viewing preference that was most pronounced for the first few fixations. In addition, her gaze fixation patterns were substantially distinct from those of controls, especially for pictures with a face. These data demonstrate an association between the clustering of neurons in face selective patches and a behavioral bias for faces in natural images.

Representations of regular and irregular shapes by deep Convolutional Neural Networks, monkey inferotemporal neurons and human judgments.

Recent studies suggest that deep Convolutional Neural Network (CNN) models show higher representational similarity, compared to any other existing object recognition models, with macaque inferior temporal (IT) cortical responses, human ventral stream fMRI activations and human object recognition. These studies employed natural images of objects. A long research tradition employed abstract shapes to probe the selectivity of IT neurons. If CNN models provide a realistic model of IT responses, then they should capture the IT selectivity for such shapes. Here, we compare the activations of CNN units to a stimulus set of 2D regular and irregular shapes with the response selectivity of macaque IT neurons and with human similarity judgements. The shape set consisted of regular shapes that differed in nonaccidental properties, and irregular, asymmetrical shapes with curved or straight boundaries. We found that deep CNNs (Alexnet, VGG-16 and VGG-19) that were trained to classify natural images show response modulations to these shapes that were similar to those of IT neurons. Untrained CNNs with the same architecture than trained CNNs, but with random weights, demonstrated a poorer similarity than CNNs trained in classification. The difference between the trained and untrained CNNs emerged at the deep convolutional layers, where the similarity between the shape-related response modulations of IT neurons and the trained CNNs was high. Unlike IT neurons, human similarity judgements of the same shapes correlated best with the last layers of the trained CNNs. In particular, these deepest layers showed an enhanced sensitivity for straight versus curved irregular shapes, similar to that shown in human shape judgments. In conclusion, the representations of abstract shape similarity are highly comparable between macaque IT neurons and deep convolutional layers of CNNs that were trained to classify natural images, while human shape similarity judgments correlate better with the deepest layers.

Neural Representations of Natural and Scrambled Movies Progressively Change from Rat Striate to Temporal Cortex.

In recent years, the rodent has come forward as a candidate model for investigating higher level visual abilities such as object vision. This view has been backed up substantially by evidence from behavioral studies that show rats can be trained to express visual object recognition and categorization capabilities. However, almost no studies have investigated the functional properties of rodent extrastriate visual cortex using stimuli that target object vision, leaving a gap compared with the primate literature. Therefore, we recorded single-neuron responses along a proposed ventral pathway in rat visual cortex to investigate hallmarks of primate neural object representations such as preference for intact versus scrambled stimuli and category-selectivity. We presented natural movies containing a rat or no rat as well as their phase-scrambled versions. Population analyses showed increased dissociation in representations of natural versus scrambled stimuli along the targeted stream, but without a clear preference for natural stimuli. Along the measured cortical hierarchy the neural response seemed to be driven increasingly by features that are not V1-like and destroyed by phase-scrambling. However, there was no evidence for category selectivity for the rat versus nonrat distinction. Together, these findings provide insights about differences and commonalities between rodent and primate visual cortex.

Visual categorization of natural movies by rats.

Visual categorization of complex, natural stimuli has been studied for some time in human and nonhuman primates. Recent interest in the rodent as a model for visual perception, including higher-level functional specialization, leads to the question of how rodents would perform on a categorization task using natural stimuli. To answer this question, rats were trained in a two-alternative forced choice task to discriminate movies containing rats from movies containing other objects and from scrambled movies (ordinate-level categorization). Subsequently, transfer to novel, previously unseen stimuli was tested, followed by a series of control probes. The results show that the animals are capable of acquiring a decision rule by abstracting common features from natural movies to generalize categorization to new stimuli. Control probes demonstrate that they did not use single low-level features, such as motion energy or (local) luminance. Significant generalization was even present with stationary snapshots from untrained movies. The variability within and between training and test stimuli, the complexity of natural movies, and the control experiments and analyses all suggest that a more high-level rule based on more complex stimulus features than local luminance-based cues was used to classify the novel stimuli. In conclusion, natural stimuli can be used to probe ordinate-level categorization in rats.

The importance of contrast features in rat vision.

Models of object recognition have mostly focused upon the hierarchical processing of objects from local edges up to more complex shape features. An alternative strategy that might be involved in pattern recognition centres around coarse-level contrast features. In humans and monkeys, the use of such features is most documented in the domain of face perception. Given prior suggestions that, generally, rodents might rely upon contrast features for object recognition, we hypothesized that they would pick up the typical contrast features relevant for face detection. We trained rats in a face-nonface categorization task with stimuli previously used in computer vision and tested for generalization with new, unseen stimuli by including manipulations of the presence and strength of a range of contrast features previously identified to be relevant for face detection. Although overall generalization performance was low, it was significantly modulated by contrast features. A model taking into account the summed strength of contrast features predicted the variation in accuracy across stimuli. Finally, with deep neural networks, we further investigated and quantified the performance and representations of the animals. The findings suggest that rat behaviour in visual pattern recognition tasks is partially explained by contrast feature processing.

When the whole is only the parts: non-holistic object parts predominate face-cell responses to illusory faces.

Humans are inclined to perceive faces in everyday objects with a face-like configuration. This illusion, known as face pareidolia, is often attributed to a specialized network of 'face cells' in primates. We found that face cells in macaque inferotemporal cortex responded selectively to pareidolia images, but this selectivity did not require a holistic, face-like configuration, nor did it encode human faceness ratings. Instead, it was driven mostly by isolated object parts that are perceived as eyes only within a face-like context. These object parts lack usual characteristics of primate eyes, pointing to the role of lower-level features. Our results suggest that face-cell responses are dominated by local, generic features, unlike primate visual perception, which requires holistic information. These findings caution against interpreting neural activity through the lens of human perception. Doing so could impose human perceptual biases, like seeing faces where none exist, onto our understanding of neural activity.

A computationally informed comparison between the strategies of rodents and humans in visual object recognition.

Many species are able to recognize objects, but it has been proven difficult to pinpoint and compare how different species solve this task. Recent research suggested to combine computational and animal modelling in order to obtain a more systematic understanding of task complexity and compare strategies between species. In this study, we created a large multidimensional stimulus set and designed a visual discrimination task partially based upon modelling with a convolutional deep neural network (CNN). Experiments included rats (N = 11; 1115 daily sessions in total for all rats together) and humans (N = 45). Each species was able to master the task and generalize to a variety of new images. Nevertheless, rats and humans showed very little convergence in terms of which object pairs were associated with high and low performance, suggesting the use of different strategies. There was an interaction between species and whether stimulus pairs favoured early or late processing in a CNN. A direct comparison with CNN representations and visual feature analyses revealed that rat performance was best captured by late convolutional layers and partially by visual features such as brightness and pixel-level similarity, while human performance related more to the higher-up fully connected layers. These findings highlight the additional value of using a computational approach for the design of object recognition tasks. Overall, this computationally informed investigation of object recognition behaviour reveals a strong discrepancy in strategies between rodent and human vision.

The neural code for "face cells" is not face-specific.

Face cells are neurons that respond more to faces than to non-face objects. They are found in clusters in the inferotemporal cortex, thought to process faces specifically, and, hence, studied using faces almost exclusively. Analyzing neural responses in and around macaque face patches to hundreds of objects, we found graded response profiles for non-face objects that predicted the degree of face selectivity and provided information on face-cell tuning beyond that from actual faces. This relationship between non-face and face responses was not predicted by color and simple shape properties but by information encoded in deep neural networks trained on general objects rather than face classification. These findings contradict the long-standing assumption that face versus non-face selectivity emerges from face-specific features and challenge the practice of focusing on only the most effective stimulus. They provide evidence instead that category-selective neurons are best understood by their tuning directions in a domain-general object space.

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.

Adaptation can explain evidence for encoding of probabilistic information in macaque inferior temporal cortex.

In predictive coding theory, the brain is conceptualized as a prediction machine that constantly constructs and updates expectations of the sensory environment [1]. In the context of this theory, Bell et al.[2] recently studied the effect of the probability of task-relevant stimuli on the activity of macaque inferior temporal (IT) neurons and observed a reduced population response to expected faces in face-selective neurons. They concluded that "IT neurons encode long-term, latent probabilistic information about stimulus occurrence", supporting predictive coding. They manipulated expectation by the frequency of face versus fruit stimuli in blocks of trials. With such a design, stimulus repetition is confounded with expectation. As previous studies showed that IT neurons decrease their response with repetition [3], such adaptation (or repetition suppression), instead of expectation suppression as assumed by the authors, could explain their effects. The authors attempted to control for this alternative interpretation with a multiple regression approach. Here we show by using simulation that adaptation can still masquerade as expectation effects reported in [2]. Further, the results from the regression model used for most analyses cannot be trusted, because the model is not uniquely defined.

Recent Visual Experience Shapes Visual Processing in Rats through Stimulus-Specific Adaptation and Response Enhancement.

From an ecological point of view, it is generally suggested that the main goal of vision in rats and mice is navigation and (aerial) predator evasion [1-3]. The latter requires fast and accurate detection of a change in the visual environment. An outstanding question is whether there are mechanisms in the rodent visual system that would support and facilitate visual change detection. An experimental protocol frequently used to investigate change detection in humans is the oddball paradigm, in which a rare, unexpected stimulus is presented in a train of stimulus repetitions [4]. A popular "predictive coding" theory of cortical responses states that neural responses should decrease for expected sensory input and increase for unexpected input [5, 6]. Despite evidence for response suppression and enhancement in noninvasive scalp recordings in humans with this paradigm [7, 8], it has proven challenging to observe both phenomena in invasive action potential recordings in other animals [9-11]. During a visual oddball experiment, we recorded multi-unit spiking activity in rat primary visual cortex (V1) and latero-intermediate area (LI), which is a higher area of the rodent ventral visual stream. In rat V1, there was only evidence for response suppression related to stimulus-specific adaptation, and not for response enhancement. However, higher up in area LI, spiking activity showed clear surprise-based response enhancement in addition to stimulus-specific adaptation. These results show that neural responses along the rat ventral visual stream become increasingly sensitive to changes in the visual environment, suggesting a system specialized in the detection of unexpected events.

Distinct and simultaneously active plasticity mechanisms in mouse hippocampus during different phases of Morris water maze training.

Although the Morris water maze (MWM) is the most frequently used protocol to examine hippocampus-dependent learning in mice, not much is known about the spatio-temporal dynamics of underlying plasticity processes. Here, we studied molecular and cellular hippocampal plasticity mechanisms during early and late phases of spatial learning in the MWM. Quantitative in situ hybridization for the immediate early genes zif268 and Homer1a (H1a) revealed phase-dependent differences in their expression between areas CA1 and CA3. During the initial learning phase, CA1 expression levels of the molecular plasticity marker H1a, but not of the activity reporter gene zif268, were related to task proficiency; whereas no learning-specific changes could be detected in CA3. Simultaneously, the ratio of surface-expressed NMDAR subunits NR2A and NR2B was downregulated as measured by acute slice biotinylation assay, while the total number of surface NMDARs was unaltered. When intrinsic 'somatic' and synaptic plasticity in the CA1-region of hippocampal slices were examined, we found that early learning promotes intrinsic neuronal plasticity as manifested by a reduction of spike frequency adaptation and postburst afterhyperpolarization. At the synaptic level, however, maintenance of long-term potentiation (LTP) in all learning groups was impaired which is most likely due to 'intrinsic' learning-induced LTP which occluded any further electrically induced LTP. Late learning, in contrast, was characterized by re-normalized H1a, NR2A and NR2B expression and neuronal firing, yet a further strengthening of learning-induced LTP. Together, our data support a precisely timed cascade of complex molecular and subcellular transformations occurring from early to late MWM learning.