The durability and specificity of perceptual learning

September, 2011

Increasing evidence shows that perception is nowhere near the simple bottom-up process we once thought. Two recent perception studies add to the evidence.

Previous research has found practice improves your ability at distinguishing visual images that vary along one dimension, and that this learning is specific to the visual images you train on and quite durable. A new study extends the finding to more natural stimuli that vary on multiple dimensions.

In the small study, 9 participants learned to identify faces and 6 participants learned to identify “textures” (noise patterns) over the course of two hour-long sessions of 840 trials (consecutive days). Faces were cropped to show only internal features and only shown briefly, so this was not a particularly easy task. Participants were then tested over a year later (range: 10-18 months; average 13 and 15 months, respectively).

On the test, participants were shown both images from training and new images that closely resembled them. While accuracy rates were high for the original images, they plummeted for the very similar new images, indicating that despite the length of time since they had seen the original images, they still retained much of the memory of them.

Although practice improved performance across nearly all items and for all people, there were significant differences between both participants and individual stimuli. More interestingly, individual differences (in both stimuli and people) were stable across sessions (e.g., if you were third-best on day 1, you were probably third-best on day 2 too, even though you were doing better). In other words, learning didn’t produce any qualitative changes in the representations of different items — practice had nearly the same effect on all; differences were rooted in initial difficulty of discriminating the pattern.

However, while it’s true that individual differences were stable, that doesn’t mean that every person improved their performance the exact same amount with the same amount of practice. Interestingly (and this is just from my eye-ball examination of the graphs), it looks like there was more individual variation among the group looking at noise patterns. This isn’t surprising. We all have a lot of experience discriminating faces; we’re all experts. This isn’t the case with the textures. For these, people had to ‘catch on’ to the features that were useful in discriminating patterns. You would expect more variability between people in how long it takes to work out a strategy, and how good that strategy is. Interestingly, three of the six people in the texture group actually performed better on the test than they had done on the second day of training, over a year ago. For the other three, and all nine of those in the face group, test performance was worse than it had been on the second day of training (but decidedly better than the first day).

The durability and specificity of this perceptual learning, the researchers point out, resembles that found in implicit memory and some types of sensory adaptation. It also indicates that such perceptual learning is not limited, as has been thought, to changes early in the visual pathway, but produces changes in a wider network of cortical neurons, particularly in the inferior temporal cortex.

The second, unrelated, study also bears on this issue of specificity.

We look at a scene and extract the general features — a crowd of people, violently riotous or riotously happy? — or we look at a scene and extract specific features that over time we use to build patterns about what goes with what. The first is called “statistical summary perception”; the second “statistical learning”.

A study designed to disentangle these two processes found that you can only do one or other; you can’t derive both types of information at the same time. Thus, when people were shown grids of lines slanted to varying degrees, they could either assess whether the lines were generally leaning to the left or right, or they could learn to recognize pairs of lines that had been hidden repeatedly in the grids — but they couldn’t do both.

The fact that each of these tasks interfered with the other suggests that the two processes are fundamentally related.


Related News

There is a pervasive myth that every detail of every experience we've ever had is recorded in memory. It is interesting to note therefore, that even very familiar objects, such as coins, are rarely remembered in accurate detail1.


Subscribe to Latest newsSubscribe to Latest newsSubscribe to Latest health newsSubscribe to Latest news