The structure and capacity of visual long-term memory

We have investigated the capacity of long-term memory for visual information, and, in particular, what kinds of background knowledge might support such a capacity. We have found evidence that long-term memory can not only store thousands of objects, but also can store those objects with a remarkable amount of visual detail. We have found that this capacity seems to be mediated by our knowledge about these objects: the more conceptually distinct the objects, the better we are able to remember large numbers of them. We've recently expanded these results into scene memory, and have begun to examine the bigger issues of exactly what makes a given set of information easy or hard to remember.

Dependence between items in visual working memory

When we look at the world, we have prior knowledge about the things we see. How does such knowledge affect the amount of visual information we can hold in mind at once? Nearly all measures of working memory capacity have attempted to quantify how many items observers can remember, and assumed that items are represented independently. By contrast, we have investigated working memory capacity from a constructive memory perspective: Rather than assuming observers' representations are based on the storage of independent items, we have shown that observers take advantage of prior knowledge when representing a display, and represent the display hierarchically by encoding a summary of the display in addition to item-level information.

Statistical learning

We've investigated our ability to learn statistical regularities from the world under two main headings. The first is visual search: We frequently search for objects during our everyday lives, and the context we find ourselves in helps predict where the object we are searching for will be located. We've investigated what information we extract and remember as we perform visual search tasks, particularly what we learn by performing a search over and over again in the same context (contextual cueing), and what we learn as we perform a particular search (rapid resumption).

In addition, we've investigated the ability to learn more high-level regularities. For example, if I leave my office and enter another room, the chance that the new room is a zoo is nearly zero, whereas the chance that it is a corridor is extremely high. Our work has shown that statistical learning can occur at this categorical level as well. Learning at this level probably helps us compress the amount of information we need to store, since any regularity we learn applies to many possible exemplars. Using learned statistical regularities for compression is also evident in my work on how statistical learning affects working memory capacity.

Perceptual organization across spatial scales

Visual perception and recognition are problems of induction - problems where we are given ambiguous input and must decide which of many possible interpretations to take. In middle-level vision, this is typically referred to as the problem of perceptual organization: how we take the bits and pieces of visual information that are available in the retinal image and structure them into larger units like objects. We have investigated perceptual organization across different spatial frequencies - how the blurry, low spatial frequency of an image and the fine, high spatial frequency details in that image interact to form our eventual percept. Our visual system is thought to break down images by spatial frequency early on in the visual pathway, so examining perceptual organization across spatial frequencies allows us to get a better idea of the types of integration the visual system has to deal with as it builds a representation of the world. We have concluded that assymmetric hysteresis effects allow our visual system to integrate across spatial frequencies in a way that provides us with the most accurate interpretation of the world as we move through it.