Digital embryos: a novel technical approach to investigate perceptual categorization in pigeons (Columba livia) using machine learning

Abstract

Pigeons are classic model animals to study perceptual category learning. To achieve a deeper understanding of the cognitive mechanisms of categorization, a careful consideration of the employed stimulus material and a thorough analysis of the choice behavior is mandatory. In the present study, we combined the use of "virtual phylogenesis", an evolutionary algorithm to generate artificial yet naturalistic stimuli termed digital embryos and a machine learning approach on the pigeons' pecking responses to gain insight into the underlying categorization strategies of the animals. In a forced-choice procedure, pigeons learned to categorize these stimuli and transferred their knowledge successfully to novel exemplars. We used peck tracking to identify where on the stimulus the animals pecked and further investigated whether this behavior was indicative of the pigeon's choice. Going beyond the classical analysis of the binary choice, we were able to predict the presented stimulus class based on pecking location using a k-nearest neighbor classifier, indicating that pecks are related to features of interest. By analyzing error trials with this approach, we further identified potential strategies of the pigeons to discriminate between stimulus classes. These strategies remained stable during category transfer, but differed between individuals indicating that categorization learning is not limited to a single learning strategy.

Keywords: Avian; Common elements; Learning; Virtual phylogenesis; Visual system.

Fig. 1

Example trial of the categorization task. Each trial began with a short high tone. First, the initialization key was presented for up to 4 s and could be terminated once pecked. Then, one digital embryo stimulus was presented (depicted is a digital embryo out of class X) and had to be pecked at least 5 times. It could not be terminated for a fixed time interval of 4 s. Subsequently, two choice keys were presented alongside the stimulus for 4 s. One peck on either key indicated the pigeon’s choice and was either immediately rewarded with 2 s access to grain or punished with lights turned off and a low punishment tone for 2 s. For half of the pigeons, the right choice key represented class X and the left choice key represented class Y (vice versa for the other half of the pigeons). The next trial followed after a 6 s inter-trial interval

Fig. 2

Digital embryo classes. Digital embryos were created with a process called “virtual phylogenesis” that mimics biological evolution. Starting with an icosahedron, different generations (G₀–G₂) are created, resulting in embryo classes X and Y

Fig. 3

Rationale of the heatmap analysis exemplified for pigeon 578. In row A, the peck distribution for correctly responded stimulus presentations is shown separately for stimulus class X and Y. To create heatmaps, the stimulus display was divided in sections of 15 × 15 squares and pecks were counted in each section. Heatmaps were created based on relative pecks to compare individual animals. The subtraction of both heatmaps reveals the different locations of pecks for each stimulus class. The precise location of each peck is used by the classifier to separate both classes resulting in the correct–correct (CC) classification. Row B shows the peck distribution for incorrectly responded stimulus presentations. The subtraction of both heatmaps reveals the different locations of pecks for each stimulus class in error trials. The precise location of each peck is used by the classifier to separate both classes resulting in the error–error (EE) classification. Row C depicts the subtraction of correct and incorrect heatmaps for each class separately resulting in the correct–error (CE) classification. The depicted analysis is exemplary based on the responses of pigeon 578 (the symbol above the identifier number indicates this individual pigeon throughout all figures). Response profiles of each individual pigeon are given in supplementary Figs. 4–10

Fig. 4

Behavioral results. Violin plot of behavioral performance for known and transfer trials overall and broken down by embryo class. Digital embryos could be categorized in each experimental condition under non-differential reward conditions. There was no difference between the performance to the known embryos and the transfer to new instances of embryo classes X and Y in any of both conditions. The symbols represent the performance of each individual pigeon and apply throughout all figures. Boxplots represent the lower quartile (Q1), the median and the upper quartile (Q3). Whiskers represent Q1 − 1.5 * interquartile range (IQR) and Q3 + 1.5 * IQR

Fig. 5

CC classification results. Digital embryos could be classified for seven out of eight animals tested. Classifier responses for the known stimuli are shown in dark gray and the classifier results for the transfer stimuli are depicted in light gray. Boxplots represent the lower quartile (Q1), the median and the upper quartile (Q3). Whiskers represent Q1 − 1.5 * IQR and Q3 + 1.5 * IQR

Fig. 6

Heatmap analysis for all individual pigeons. Only pecks from correct trials for the two stimulus classes (left: X, right: Y) are shown. Additional analyses as presented in Fig. 3 are given in the supplementary figures

Fig. 7

Single peck analysis. The classifier response for the known stimuli is depicted for each animal and for each of the five consecutive pecks. For the majority of the animals, the classification accuracy increases for consecutive pecks. Boxplots represent the lower quartile (Q1), the median and the upper quartile (Q3). Whiskers represent Q1 − 1.5 * IQR and Q3 + 1.5 * IQR

Fig. 8

CE (correct–error) classification results of the second experimental stage. A shows CE classification for the known stimuli. B depicts the CE classification for the transfer stimuli. Dark red colors indicate that the pigeons showed the “wrong” behavior for the respective stimulus class and thus a confusion between the categories X and Y. Light red colors indicate that the animals showed the “right” behavior for the respective stimulus class resulting in a confusion within a given category in the CE classification. Boxplots represent the lower quartile (Q1), the median and the upper quartile (Q3). Whiskers represent Q1 − 1.5 * IQR and Q3 + 1.5 * IQR (color figure online)