Abstract-concept learning of difference in pigeons

Exploring Higher-Order Conceptual Learning in an Arthropod with a Large Multisensory Processing Center

Simple Summary

It is difficult to measure animal intelligence because the definition of ‘intelligence’ varies, and many animals are good at specific tasks used to measure intelligence or cognition. To address this, scientists often look for evidence of common cognitive abilities. One such ability, the ability to learn concepts, is thought to be rare in animals, especially invertebrates. Concepts include the ideas of ‘same’ and ‘different’. These concepts can be applied to anything in the environment while also being independent of those objects and can help animals understand and survive their environment. Amblypygids, a relative of spiders, live in tropical and subtropical areas, are very good learners, and have a large, complex brain region known to process information from multiple senses. We tested whether amblypygids could learn the concept of ‘same’ by training them to move toward a stimulus that matched with an initial stimulus. We also trained some individuals to learn the concept ‘different’ by training them to move toward a non-matching stimulus. When we used new stimuli, the amblypygids did not move toward the correct stimulus significantly more often than the incorrect stimulus, suggesting either they are unable to learn these higher-order concepts or our experimental design failed to elicit that ability.

Abstract

Comparative cognition aims to understand the evolutionary history and current function of cognitive abilities in a variety of species with diverse natural histories. One characteristic often attributed to higher cognitive abilities is higher-order conceptual learning, such as the ability to learn concepts independent of stimuli—e.g., ‘same’ or ‘different’. Conceptual learning has been documented in honeybees and a number of vertebrates. Amblypygids, nocturnal enigmatic arachnids, are good candidates for higher-order learning because they are excellent associational learners, exceptional navigators, and they have large, highly folded mushroom bodies, which are brain regions known to be involved in learning and memory in insects. In Experiment 1, we investigate if the amblypygid Phrynus marginimaculatus can learn the concept of same with a delayed odor matching task. In Experiment 2, we test if Paraphrynus laevifrons can learn same/different with delayed tactile matching and nonmatching tasks before testing if they can transfer this learning to a novel cross-modal odor stimulus. Our data provide no evidence of conceptual learning in amblypygids, but more solid conclusions will require the use of alternative experimental designs to ensure our negative results are not simply a consequence of the designs we employed.

Differential Involvement of EEG Oscillatory Components in Sameness versus Spatial-Relation Visual Reasoning Tasks

The development of deep convolutional neural networks (CNNs) has recently led to great successes in computer vision, and CNNs have become de facto computational models of vision. However, a growing body of work suggests that they exhibit critical limitations on tasks beyond image categorization. Here, we study one such fundamental limitation, concerning the judgment of whether two simultaneously presented items are the same or different (SD) compared with a baseline assessment of their spatial relationship (SR). In both human subjects and artificial neural networks, we test the prediction that SD tasks recruit additional cortical mechanisms which underlie critical aspects of visual cognition that are not explained by current computational models. We thus recorded electroencephalography (EEG) signals from human participants engaged in the same tasks as the computational models. Importantly, in humans the two tasks were matched in terms of difficulty by an adaptive psychometric procedure; yet, on top of a modulation of evoked potentials (EPs), our results revealed higher activity in the low β (16–24 Hz) band in the SD compared with the SR conditions. We surmise that these oscillations reflect the crucial involvement of additional mechanisms, such as working memory and attention, which are missing in current feed-forward CNNs.

Towards describing scenes by animals: Pigeons' ordinal discrimination of objects varying in depth

The perception of a complex scene requires visual mechanisms that include identifying objects and their relative placement in depth. To examine apparent depth perception in birds, we tested four pigeons with a novel multiple-sequential-choice procedure. We created 3D-rendered scene stimuli containing three objects located at different apparent depths based on a variety of pictorial cues and placed small circular target response areas on them. The pigeons were trained to sequentially choose among the multiple response areas to report the object closest in apparent depth (ordinal position; front then middle object). After the pigeons learned this sequential depth discrimination, their use of three different monocular depth cues (occlusion, relative size, height in field) was tested, and their flexibility evaluated using three novel objects. In addition to the contribution to understanding apparent depth perception in birds, the use of more flexible open-ended choice discriminations, as employed here, has considerable promise for creating informative production-like tasks in nonverbal animals.

Effects of set size on identity and oddity abstract-concept learning in rats

Match (MTS) and non-match-to-sample (NMTS) procedures are used to assess concepts of identity and oddity across species and are measured by transfer performance to novel stimuli. The number of exemplars used in training (set size) has been shown to affect learning with evidence of larger set sizes promoting concept learning in several species. The present study explored the effects of set size and procedure on concept learning in rats using olfactory stimuli. Concept learning was assessed for 20 rats via transfer tests consisting of novel stimuli after rats were initially trained to either MTS or NMTS with two or ten stimuli as exemplars. No difference was found in acquisition or transfer between MTS and NMTS, but rats trained with ten stimuli performed better on novel transfer tests than rats trained with two. When set size was expanded for rats originally trained with two stimuli and rats were re-tested with ten novel stimuli, performance showed full transfer demonstrating that training with multiple exemplars facilitates concept learning.

Not-So-CLEVR: learning same-different relations strains feedforward neural networks

The advent of deep learning has recently led to great successes in various engineering applications. As a prime example, convolutional neural networks, a type of feedforward neural network, now approach human accuracy on visual recognition tasks like image classification and face recognition. However, here we will show that feedforward neural networks struggle to learn abstract visual relations that are effortlessly recognized by non-human primates, birds, rodents and even insects. We systematically study the ability of feedforward neural networks to learn to recognize a variety of visual relations and demonstrate that same-different visual relations pose a particular strain on these networks. Networks fail to learn same-different visual relations when stimulus variability makes rote memorization difficult. Further, we show that learning same-different problems becomes trivial for a feedforward network that is fed with perceptually grouped stimuli. This demonstration and the comparative success of biological vision in learning visual relations suggests that feedback mechanisms such as attention, working memory and perceptual grouping may be the key components underlying human-level abstract visual reasoning.

Abstraction, Multiple Exemplar Training and the Search for Derived Stimulus Relations in Animals

Symmetry and other derived stimulus relations are readily demonstrated in humans in a variety of experimental preparations. Comparable emergent relations are more difficult to obtain in other animal species and seem to require certain specialized conditions of training and testing. This article examines some of these conditions with an emphasis on what animal research may be able to tell us about the nature and origins of derived stimulus relations. We focus on two areas that seem most promising: 1) research generated by Urcuioli's (2008) theory of the conditions necessary to produce symmetry in pigeons, and 2) research that explores the effects of multiple exemplar training on emergent relations. Urcuioli's theory has successfully predicted emergent relations in pigeons by taking into account their apparent difficulty in abstracting the nominal training stimulus from other stimulus properties such as location and temporal position. Further, whereas multiple exemplar training in non-humans has not consistently yielded arbitrarily-applicable relational responding, there is a growing body of literature showing that it does result in abstracted same-different responding. Our review suggests that although emergent stimulus relations demonstrated in non-humans at present have not yet shown the flexibility or generativity apparent in humans, the research strategies reviewed here provide techniques that may permit the analysis of the origins of derived relational responding.

Concurrent learning of multiple oddity discrimination in rats

This experimental series examines rats' concurrent learning of multiple oddity discrimination and its transfer to novel stimuli. Three rats were trained to discriminate an odd object from five identical objects in a row. After acquisition of the AAAAAB problem, the BBBBBA set was added to training. At the start of concurrent training with both sets, performance on BBBBBA was significantly below chance, suggesting that rats had acquired a tendency to respond to and/or avoid specific features of objects during the initial AAAAAB training. Although all rats learned the two problems reliably, no transfer effect was observed during tests with novel sets CCCCCD and DDDDDC. After the first transfer test, rats performed reliably for stimulus sets AAAAAB, BBBBBA, CCCCCD, and DDDDDC concurrently. Although one rat showed reliable transfer for novel test problems EEEEEF and FFFFFE, the possibility of biased test performance between stimulus sets, by choosing one of two test objects, could not be excluded. However, following the transfer tests, all rats responded significantly to 20 novel problems immediately after they were introduced into training. These findings offer evidence of rats' capacity for concurrent oddity discrimination using multiple stimulus sets, as well as preliminary evidence of relational oddity learning in rats.

A negative stimulus movement effect in pigeons

Rhesus monkeys and humans perform more accurately in matching-to-sample tasks when the sample stimulus moves through space (Washburn et al., 1989; Washburn, 1993). This Stimulus Movement Effect (SME) is believed to be due to movement increasing attention toward the sample stimulus, creating an easier discrimination between the sample and choice stimuli. To date, there is no evidence for this phenomenon in a non-mammalian species. In the current study, we investigate the possibility of an SME in an avian species. Across three experiments, pigeons were tested with moving and stationary sample stimuli in a non-matching- to-sample task. The area and velocity by which the sample stimulus traveled was manipulated but no advantage for moving over stationary sample trials was found within or across sessions. Even when a delay condition was implemented, there was no advantage for moving sample trials. Contrary to the results found in humans and monkeys, pigeons performed better when the sample was stationary, a negative SME, and no evidence was found that stimulus movement increases discrimination performance.

Emergent identity but not symmetry following successive olfactory discrimination training in rats

The search for symmetry in nonhuman subjects has been successful in recent studies in pigeons (e.g., Urcuioli, 2008). The key to these successes has been the use of successive discrimination procedures and combined training on identity, as well as arbitrary, baseline relations. The present study was an effort to extend the findings and theoretical analysis developed by Urcuioli and his colleagues to rats using olfactory rather than visual stimuli. Experiment 1 was a systematic replication of Urcuioli's (2008) demonstration of symmetry in pigeons. Rats were exposed to unreinforced symmetry probes following training with two arbitrary and four identity conditional discriminations. Response rates on symmetry probe trials were low and provided little evidence for emergent symmetry in any of the seven rats tested. In Experiment 2, a separate group of six rats was trained on four identity relations and was then exposed to probe trials with four novel odor stimuli. Response rates were high on identity probe trials, and low on nonmatching probe trials. The similar patterns of responding on baseline and probe trials that were shown by most rats provided a demonstration of generalized identity matching. These findings suggest that the development of stimulus control topographies in rats with olfactory stimuli may differ from those that emerge in pigeons with visual stimuli. Urcuioli's (2008) theory has been highly successful in predicting conditions necessary for stimulus class formation in pigeons, but may not be sufficient to fully understand determinants of emergent behaviors in other nonhuman species.

Temporal dynamics of task switching and abstract-concept learning in pigeons

The current study examined whether pigeons could learn to use abstract concepts as the basis for conditionally switching behavior as a function of time. Using a mid-session reversal task, experienced pigeons were trained to switch from matching-to-sample (MTS) to non-matching-to-sample (NMTS) conditional discriminations within a session. One group had prior training with MTS, while the other had prior training with NMTS. Over training, stimulus set size was progressively doubled from 3 to 6 to 12 stimuli to promote abstract concept development. Prior experience had an effect on the initial learning at each of the set sizes but by the end of training there were no group differences, as both groups showed similar within-session linear matching functions. After acquiring the 12-item set, abstract-concept learning was tested by placing novel stimuli at the beginning and end of a test session. Prior matching and non-matching experience affected transfer behavior. The matching experienced group transferred to novel stimuli in both the matching and non-matching portion of the sessions using a matching rule. The non-matching experienced group transferred to novel stimuli in both portions of the session using a non-matching rule. The representations used as the basis for mid-session reversal of the conditional discrimination behaviors and subsequent transfer behavior appears to have different temporal sources. The implications for the flexibility and organization of complex behaviors are considered.