Visual object recognition in primates

The ability of human participants and rhesus monkeys to correctly categorise large numbers of visual images has been examined in several recent experiments..

The abstracts of the papers are given below, and also some brief notes about the experimental procedures in each case.

Fabre-Thorpe, M. (2003). Visual categorization: accessing abstraction in non-human primates. Philosophical Transactions of the Royal Society of London Series B-Biological Sciences, 358(1435), 1215-1223.

Evolution might have set the basic foundations for abstract mental representation long ago. Because of language, mental abilities would have reached different degrees of sophistication in mammals and in humans but would be, essentially, of the same nature. Thus, humans and animals might rely on the same basic mechanisms that could be masked in humans by the use of sophisticated strategies. In this paper, monkey and human abilities are compared in a variety of perceptual tasks including visual categorization to assess behavioural similarities and dissimilarities, and to determine the level of abstraction of monkeys' mental representations. The question of how these abstract representations might be encoded in the brain is then addressed. A comparative study of the neural processing underlying abstract cognitive operations in animals and humans might help to understand when abstraction emerged in the phylogenetic scale, and how it increased in complexity.

Brief notes: The Fabre-Thorpe protocol (see the Delorme paper.) had images projected onto touch screen 30 cm away for only 32 ms. There was a random interval 1.5-3.0 sec between image trials. Subjects had to touch a key below the screen to begin a trial, then they had 1 sec to transfer their hand to touch the screen if they detected a target (animal or food), otherwise they had to keep their hand on the key below the screen. At end of the 1 sec there was a beep for correct, and a drop of orange juice for the monkeys. If the response made was wrong there was a 3-4 sec display of the wrongly classified picture. Monkeys sat in a primate chair and worked daily for as long as they wished (1-3 hours) 5 days per week. www.pdf version of this paper (log on first for access outside the College.)

Freedman, D. J., Riesenhuber, M., Poggio, T., & Miller, E. K. (2001). Categorical representation of visual stimuli in the primate prefrontal cortex. Science, 291(5502), 312-316.

The ability to group stimuli into meaningful categories is a fundamental cognitive process. To explore its neural basis. we trained monkeys to categorize computer-generated stimuli as "cats" and "dogs." A morphing system was used to systematically vary stimulus shape and precisely define the category boundary. Neural activity in the Lateral prefrontal cortex reflected the category of visual stimuli, even when a monkey was retrained with the stimuli assigned to new categories.

The Freedman et al paper using morphed stimuli involved a matching procedure with pictures presented successively. To start a trial the monkey had to grasp metal bar and fixate the centre of the screen for 500 ms. The a sample picture (either "dog" or "cat") was presented for 600 ms, followed after a 1 sec delay by the choice image. If the choice image was the same category as the sample l second previous then the monkey needed to release the lever within the 600 ms display time in order to be rewarded with juice. If choice image was in the different category then the animal was required to keep gripping its lever for the 600 ms and then a stimulus would appear which was in the same category and the monkey could then release the bar to be rewarded with juice. There were intervals of 2 to 3 secs between trials. For an error an additional 3 second time- out occurred. The animals typically achieved about 700 correct trials per day and were more that 90% correct. Training started with just the 6 "prototypes" and the choice stimuli being identical to the sample. During the training more than 1000 stimuli were used from all over the morph space. But for the purpose recording from the prefrontal neurons only 54 sample images were used and for the choice stimuli 6 levels of blend between cat and dog were in 100 images otherwise chosen at random. www.pdf version of this paper (from author's home page)

Sigala, N., Gabbiani, F., & Logothetis, N. K. (2002). Visual categorization and object representation in monkeys and humans. Journal of Cognitive Neuroscience, 14(2), 187-198.

We investigated the influence of a categorization task on the extraction and representation of perceptual features in humans and monkeys. The use of parameterized stimuli (schematic faces and fish) with fixed diagnostic features in combination with a similarity-rating task allowed us to demonstrate perceptual sensitization to the diagnostic dimensions of the categorization task for the monkeys. Moreover, our results reveal important similarities between human and monkey visual subordinate categorization strategies. Neither the humans nor the monkeys compared the new stimuli to class prototypes or based their decisions on conditional probabilities along stimulus dimensions. Instead, they classified each object according to its similarity to familiar members of the alternative

The Sigala, Gaibbiana and Logothetis paper used 12 humans and 3 monkeys. The monkeys got juice rewards. They had two levers and had to pull one or the other according to the category of the presented image. Low and high tones provided feedback. After learning to be 75% correct they got blocks of trials with juice only after 5 consecutive correct responses. After 85% correct with the 10 exemplars, new exemplars were tested with no feedback given. For each subject there were 15 classifications per stimulus. They also did similarity testing For the Bruswick faces the Eye Height and Eye Separation were relevant, Nose Length and Mouth Height not. For the Fish, Dorsal Fin and Tail were relevant, and Mouth and Ventral Fin not. All dimensions had 3 possible values [see the summary of this paper by Gauthier and Palmeri](log on first for access outside the College)

 

Bovet, D., Vauclair, J., & Blaye, A. (2005). Categorization and abstraction abilities in 3-year-old children: a comparison with monkey data. Animal Cognition, 8(1), 53-59.

Three-year-old children were tested on three categorization tasks of increasing levels of abstraction (used with adult baboons in an earlier study): the first was a conceptual categorization task (food vs toys), the second a perceptual matching task (same vs different objects), and the third a relational matching task in which the children had to sort pairs according to whether or not the two items belonged to the same or different categories. The children were tested using two different procedures, the first a replication of the procedure used with the baboons (pulling one rope for a category or a relationship between two objects, and another rope for the other category or relationship), the second a task based upon children's prior experiences with sorting objects (putting in the same box objects belonging to the same category or a pair of objects exemplifying, the same relation). The children were able to solve the first task (conceptual categorization) when tested with the sorting into boxes procedure, and the second task (perceptual matching) when tested with both procedures. The children were able to master the third task (relational matching) only when the rules were clearly explained to them, but not when they could only watch sorting examples. In fact, the relational matching task without explanation requires analogy abilities that do not seem to be fully developed at 3 years of age. The discrepancies in performances between children tested with the two procedures, with the task explained or not, and the discrepancies observed between children and baboons are discussed in relation to differences between species and/or problem-solving strategies.

This paper compares the performance of 3 year old children with the findings previously obtained with adult baboons (Bovet & Vaclair, 2001). The baboons were tested with an apparatus attached to their home enclosure. Two objects were presented, and the task for the animal was to pull at a rope at one side if the objects were the same, but to pull a second rope on the other side if the two objects were different. Small food rewards were delivered for correct responses. The baboons were able to learn either if the two objects were in the same category, using the categories of "food" and "non-food". The 3-year old children found this difficult to learn unless the experimenter explained to the child “that he/she should place pictures of pairs of things that go well together in one box, because they are both for eating or both for playing, and in the other box, pictures of pairs of things that do not go well together, because one is for playing and the other is for eating”.