Learning to Categorize Objects

Many people perform societally important categorization tasks as their full-time jobs, such as airport security personnel checking baggage, and dermatologists deciding whether skin moles are cancerous. These tasks usually require lengthy training to master. In my research program, I looked at real-world-inspired category learning tasks, and examined how the training process could be improved. In Chapters 1 and 2, I had participants learn to classify objects into two categories. In Chapter 3, I had participants perform a visual search task and examined the learning effect in the task. In Chapter 1, I studied people's category learning efficacy in the case of overshadowing. Overshadowing effect occurs when an object feature, while strongly associates with a category during training, does not appear in the transfer test. Learners tend to rely on the feature in the training phase, and do poorly due to its absence in the transfer test. I examined whether overshadowing effect could be eliminated through top-down instructions. Chapter 2 examined a new type of training schedule. The literature on category learning has primarily focused on using massed or interleaved learning as the training schedules, and discovered that each has its strengths and weaknesses. Here I proposed a new way to train learners, by presenting them with exemplars from each of the two categories simultaneously within each trial. Learning efficacy of the new training schedule is comparable to that of the interleaved schedule, while greatly reduces learners' frustrations during training. I will present data using artificial and real-world stimuli. In the last chapter, I examined the effect of implicit learning during visual search. Participants were asked to search for a small set of targets among the same set of distractors repeated across trials. Three sets of distractors were manipulated across participants. After participants had been trained on a set of distractors, a new set of distractors was introduced. Search performance was worse when the target-distractor discriminability was higher in the second phase, indicating that distractor properties were learned during repeated visual search