Statistical learning of new visual feature combinations by infants

Abstract

The ability of humans to recognize a nearly unlimited number of unique visual objects must be based on a robust and efficient learning mechanism that extracts complex visual features from the environment. To determine whether statistically optimal representations of scenes are formed during early development, we used a habituation paradigm with 9-month-old infants and found that, by mere observation of multielement scenes, they become sensitive to the underlying statistical structure of those scenes. After exposure to a large number of scenes, infants paid more attention not only to element pairs that cooccurred more often as embedded elements in the scenes than other pairs, but also to pairs that had higher predictability (conditional probability) between the elements of the pair. These findings suggest that, similar to lower-level visual representations, infants learn higher-order visual features based on the statistical coherence of elements within the scenes, thereby allowing them to develop an efficient representation for further associative learning.

Figures

Fig 1.

Stimulus elements and scenes used in the experiments. (a) The twelve shapes were grouped into four base pairs and four noise elements, with each noise element appearing with only one base pair. (b) The four possible scenes created by one base pair and its noise element. In Experiment 1, all four scenes were presented during habituation; in Experiments 2 and 3, again, all four scenes were presented for low-frequency base pairs, but only two scenes (shown in the right column) were presented for high-frequency base pairs, which appeared twice as often as individual scenes with low-frequency base pairs. Because this doubling of appearance frequency was equally split between the two scenes containing the high-frequency base pair, the non-base pair (marked by the white rectangle in the lower right scene) appeared the same number of times as a low-frequency base pair.

Fig 2.

Results of Experiment 1. There was a very strong looking preference for base pairs over non-base pairs, suggesting that infants noticed the higher cooccurrence of elements within the base pairs.

Fig 3.

Relation between the appearance frequency of the base pairs and the cooccurrence and predictability of their elements. (a) In Experiment 1, all base pairs were presented an equal number of times to create the scenes (Left). Therefore, the cooccurrence of the elements of each base pair [measured by the joint probability P(a,b), Upper Right] was uniformly higher than that of the non-base pairs. In addition, this relation was identical whether measured by the cooccurrence of elements or by the predictability between the elements quantified by the conditional probability P(a,b) (Lower Right). (b) In Experiments 2 and 3, there were base pairs used with low (gray) and high (black) frequency to generate the scenes (Left). Consequently, there were differences in cooccurrence frequency and predictability within both the base pairs and the non-base pairs of the low- and high-frequency types. With appropriate selection of relative frequencies, the cooccurrence of elements within the low-frequency base pairs (Upper Right, gray bars) was equated with that of the high-frequency non-base pairs (Upper Right, black bars), whereas the predictability of elements within those two types remained significantly different (Lower Right), thereby decoupling cooccurrence frequency and predictability.

Fig 4.

(a) Results of Experiment 2. Infants had a strong looking preference for the base pairs with higher predictability between elements than for the frequency-balanced non-base pairs with lower predictability, even though the appearance frequency of elements in the frequency-balanced non-base pairs was higher by a factor of 2 for elements in the base pairs, and the cooccurrence frequency of the low-frequency base pairs and the frequency-balanced non-base pairs was equated. (b) Results of Experiment 3. Infants showed no discrimination of single elements despite their varying in appearance frequency by a factor of 2. The absence of a posthabituation preference for individual elements in this control experiment rules out element frequency as an explanation of the preference for low-frequency base pairs in Experiment 2.