Perceptual Learning: Basic Research and Modeling

In contemporary work, perceptual learning refers to experience-induced improvements in the pickup of information (Gibson, 1969). Our work in perceptual learning (PL) has several strands.

We have developed and explored the distinction between discovery and fluency effects in PL. Discovery processes involve finding and selectively extracting the information relevant to a task, including becoming sensitive to relational information that was not even encoded prior to learning. Fluency processes involve extracting the information identified by discovery processes more quickly, more in parallel, and with lower attentional load.

In recent work, we have shown the importance of perceptual learning in high-level cognitive tasks, including symbolic ones, such as mathematics learning. As Kellman & Massey (2013) wrote "Much of thinking turns out to be seeing, if seeing is properly understood." These connections require, however, revised conceptions of perception, cognition, and perceptual learning, at least with reference to common understandings of them (Kellman & Massey, 2013). A classical view, still too common, is that perception furnishes only rudimentary or sensory information. Complexities such as the spatial, temporal, causal, or animate structure of objects, situations, and events are derived from the raw materials given by perception through higher cognitive processes. In contrast, modern views of perception show that it is specialized for extraction of complex relations and meaningful structure that is directly relevant to cognitive processes. Even supposedly basic operations such as seeing the orientation of an edge or the color of a surface are results of complex computations aimed at functionally relevant descriptions of the world. Similarly, in perceptual learning, although much work in recent decades has focused on simple stimuli, we have argued that the same processes of information selection and discovery and fluent extraction of relevant relations provides a crucial basis of expertise in real-world learning domains. Some of our research has shown that even supposedly "low-level" PL depends on representations based on perceptual constancy. A great deal of work has shown the application of PL concepts to extraction of invariance from amidst variation in high-level, real-world, learning domains (see other tabs).

Finding laws of PL and deriving models of PL from our research and research by others, and using them to develop interactive learning tools that can accelerate the growth of expertise. These activities are described elsewhere on our website.

Also, in recent work we are investigating the relationship between phenomena of perceptual learning and statistical learning, concepts that are sometimes used differently and sometimes interchangeably in the research literature. Our preliminary work argues that these two forms of learning are conceptually distinguishable but may co-occur and collaborate in interesting ways that depend on learning conditions.

Perceptual Development

Our laboratory has a longstanding involvement in the development of perception in human infancy. Most recently, this is reflected in the book by Martha Arterberry and Phil Kellman Development of Perception in Infancy: The Cradle of Knowledge Revisited (Oxford University Press, 2016). This book offers a unified treatment and critical consideration of what is known about how perception develops. It updates our earlier book The Cradle of Knowledge (MIT Press, 1998).

The Cradle of Knowledge
Kellman, P.J. & Arterberry, M. (1998). The Cradle of Knowledge: Perceptual Development in Infancy. Cambridge, MA: MIT Press.
In earlier years, beginning in the 1980s, research from our laboratory helped to describe the origins of visual perception of object, space, and motion perception in human infants. The Kellman & Spelke work on infants’ perception of partly occluded objects represents some of the earliest research showing advanced perceptual abilities early in life and appears in introductory, developmental, and perception textbooks. Our lab produced the first clear demonstrations of three-dimensional (3D) form perception in human infants (Kellman, 1984; Kellman & Short, 1987) and suggested methods that have been used to study 3D form perception in animals. Much of our work involved relations between motion and object perception. We discovered that infants distinguish object from observer motion and showed that only real object motion produces perception of complete objects under occlusion. Many of our results support the general idea that much of early visual competence in humans rests on kinematic (motion) information (e.g., Kellman, 1992).

These efforts have helped to replace conceptions of the origins of perception from protracted associative learning to a view of development clearly rooted in innate and early maturing capacities that provide meaningful perception of objects, layout, and events from early life and are fine-tuned by experience. This picture has not only replaced centuries of misunderstanding about how perception develops but also revises our understanding of the perceptual foundations that are crucial for cognitive, linguistic, and social development.


  • Philip J. Kellman Philip J. Kellman
  • Christine Massey Christine Massey (UCLA)
  • Everett Mettler Everett Mettler
  • Carolyn Bufford Carolyn Bufford
  • Victoria Jacoby Victoria Jacoby


  • face Hongjing Lu (UCLA)
  • face Patrick Garrigan (SJU)
  • face Martha Arterberry (Colby College)

Selected Publications

Perceptual Learning

Perceptual Development