Teaching pigeons an abstract relational rule: insideness

Beyond the feedforward sweep: feedback computations in the visual cortex

Visual perception involves the rapid formation of a coarse image representation at the onset of visual processing, which is iteratively refined by late computational processes. These early versus late time windows approximately map onto feedforward and feedback processes, respectively. State-of-the-art convolutional neural networks, the main engine behind recent machine vision successes, are feedforward architectures. Their successes and limitations provide critical information regarding which visual tasks can be solved by purely feedforward processes and which require feedback mechanisms. We provide an overview of recent work in cognitive neuroscience and machine vision that highlights the possible role of feedback processes for both visual recognition and beyond. We conclude by discussing important open questions for future research.

A Computational Model of Dual Competition between the Basal Ganglia and the Cortex

We propose a model that includes interactions between the cortex, the basal ganglia (BG), and the thalamus based on a dual competition. We hypothesize that the striatum, the subthalamic nucleus (STN), the internal globus pallidus (GPi), the thalamus, and the cortex are involved in closed feedback loops through the hyperdirect and direct pathways. These loops support a competition process that results in the ability of BG to make a cognitive decision followed by a motor one. Considering lateral cortical interactions, another competition takes place inside the cortex allowing the latter to make a cognitive and a motor decision. We show how this dual competition endows the model with two regimes. One is driven by reinforcement learning and the other by Hebbian learning. The final decision is made according to a combination of these two mechanisms with a gradual transfer from the former to the latter. We confirmed these theoretical results on primates (Macaca mulatta) using a novel paradigm predicted by the model.

A natural history of skills

The dorsal pallium (a.k.a. cortex in mammals) makes a loop circuit with the basal ganglia and the thalamus known to control and adapt behavior but the who's who of the functional roles of these structures is still debated. Influenced by the Triune brain theory that was proposed in the early sixties, many current theories propose a hierarchical organization on the top of which stands the cortex to which the subcortical structures are subordinated. In particular, habits formation has been proposed to reflect a switch from conscious on-line control of behavior by the cortex, to a fully automated subcortical control. In this review, we propose to revalue the function of the network in light of the current experimental evidence concerning the anatomy and physiology of the basal ganglia-cortical circuits in vertebrates. We briefly review the current theories and show that they could be encompassed in a broader framework of skill learning and performance. Then, after reminding the state of the art concerning the anatomical architecture of the network and the underlying dynamic processes, we summarize the evolution of the anatomical and physiological substrate of skill learning and performance among vertebrates. We then review experimental evidence supporting for the hypothesis that the development of automatized skills relies on the BG teaching cortical circuits and is actually a late feature linked with the development of a specialized cortex or pallium that evolved in parallel in different taxa. We finally propose a minimal computational framework where this hypothesis can be explicitly implemented and tested.

Processing of above/below categorical spatial relations by baboons (Papiopapio)

Three video-formatted experiments investigated the categorization of 'above' and 'below' spatial relations in baboons (Papio papio). Using an identity matching-to-sample task, six baboons correctly matched line-dot stimuli based on the 'above' or 'below' location of the dot relative to the line (Experiment 1). Positive transfer of performance was then observed when the line-dot distance depicted in the sample stimulus differed from that of the two comparison stimuli (Experiment 2). Using a go/nogo procedure, two baboons were further trained to discriminate whether a 'B' character was displayed 'above' or 'below' a '3' character (Experiment 3). After training, a positive transfer of performance was observed with the same type of stimuli depicted in different type fonts. Altogether, these results suggest that baboons may form conceptual representations of 'above' and 'below' spatial relations, and categorize visual forms on that basis.

Evidence for a Numerosity Category that is Based on Abstract Qualities of "Few" vs. "Many" in the Bottlenose Dolphin (Tursiops truncatus)

A previous study (Kilian et al., 2003) had demonstrated that bottlenose dolphins can discriminate visual stimuli differing in numerosity. The aim of the present study was twofold: first, we sought to determine if dolphins are able to use a numerical category based on “few” vs. “many” when discriminating stimuli according to the number of their constituent patterns. Second, we aimed to extend the previously demonstrated range of numbers, thereby testing the limits of the numerical abilities of bottlenose dolphins. To this end, one adult bottlenose dolphin learned to discriminate between two simultaneously presented stimuli which varied in the number of elements they contained. After initial training, several confounding parameters were excluded to render it likely that discrimination performance indeed depended on numerosity. Subsequently, the animal was tested with new stimuli of intermediate as well as higher numbers of elements. Once discrimination had been achieved, a reversal-training on a subset of stimuli was initiated. Afterward, the subject generalized the reversal successful to new and unreinforced stimuli. Our results reveal two main findings: firstly, our data strongly suggest a magnitude and a distance effect. Thus, coding of numerical information in dolphins might follow logarithmic scaling as postulated by the Weber-Fechner law. Secondly, after learning a reversal of contingencies, the dolphin generalized the reversal successful to new and unreinforced stimuli. Thus, within the limits of a study that was conducted with a single individual, our results suggest that dolphins are able to learn and use a numerical category that is based on abstract qualities of “few” vs. “many.”

Changes in area affect figure-ground assignment in pigeons

A critical cue for figure-ground assignment in humans is area: smaller regions are more likely to be perceived as figures than are larger regions. To see if pigeons are similarly sensitive to this cue, we trained birds to report whether a target appeared on a colored figure or on a differently colored background. The initial training figure was either smaller than (Experiments 1 and 2) or the same area as (Experiment 2) the background. After training, we increased or decreased the size of the figure. When the original training shape was smaller than the background, pigeons' performance improved with smaller figures (and worsened with larger figures); when the original training shape was the same area as the background, pigeons' performance worsened when they were tested with smaller figures. A smaller figural region appeared to improve the figure-ground discrimination only when size was a relevant cue in the initial discrimination.

Figure-ground assignment in pigeons: evidence for a figural benefit

Four pigeons discriminated whether a target spot appeared on a colored figural shape or on a differently colored background by first pecking the target and then reporting its location: on the figure or the background. We recorded three dependent variables: target detection time, choice response time, and choice accuracy. The birds were faster to detect the target, to report its location, and to learn the correct response on figure trials than on background trials. Later tests suggested that the pigeons might have attended to the figural region as a whole rather than using local properties in performing the figure-background discrimination. The location of the figural region did not affect figure-ground assignment. Finally, when 4 other pigeons had to detect and peck the target without making a choice report, no figural advantage emerged in target detection time, suggesting that the birds' attention may not have been automatically summoned to the figural region.

Categorization of Above and Below Spatial Relations by Tufted Capuchin Monkeys (Cebus apella)

Using a matching-to-sample procedure, the researchers investigated tufted capuchins' (Cebus apella) ability to form categorical representations of above and below spatial relations. In Experiment 1, 5 capuchins correctly matched bar-dot stimuli on the basis of the relative above and below location of their constituent elements. The monkeys showed a positive transfer of performance both when the bar-dot distance in the two comparison stimuli differed from that of the sample and when the actual location of the matching stimulus and the nonmatching stimulus on the apparatus was modified. In Experiment 2, the researchers systematically changed the shapes of the located object (the dot) or the reference object (the horizontal bar). These manipulations did not affect the monkeys' performance. Overall, the data suggest that capuchins can form abstract, conceptual-like representations for above and below spatial relations.

Evaluation and development of a connectionist theory of configural learning

A configural theory of associative learning is described that is based on the assumption that conditioning results in associations between the unconditioned stimulus and a representation of the entire pattern of stimulation that was present prior to its delivery. Configural theory was formulated originally to account for generalization and discrimination in Pavlovian conditioning. The first part of the article demonstrates how this theory can be used to explain results from studies of overshadowing, blocking, summation, and discrimination learning. The second part of the article shows how the theory can be developed to explain a broader range of phenomena, including mediated conditioning, reinforcer devaluation effects, the differential outcomes effect, acquired equivalence, sensory preconditioning, and structural discriminations.

Levels of stimulus control: a functional approach

This paper surveys some illustrative experiments on categorization of visual stimuli by animals other than human. The results suggest a classification of categorical powers in five steps from simple discrimination to rote and open-ended categorization, to concepts and the use of abstract relations. Nonhuman animals evidently readily categorize up to the fourth level as here defined, which is the level of concepts. With difficulty, they can sometimes be induced to rise even to the level of abstract relations. It is at the level of abstract relations that a large gap opens up between human categorizations and categorization by other animals.