Preparing for Novelty with Diverse Training

Summary: This study investigated the ability to generalize acquired skills from training conditions to novel conditions, in a complex perceptual and cognitive task of luggage screening. We examined category and exemplar diversity during training for preparing learners to detect novel items during transfer. Category diversity was manipulated in terms of heterogeneity of training categories: Participants either trained with targets from one category or with targets from several categories. Exemplar diversity was manipulated between participants by presenting either a few or many exemplars for both category diversity conditions. Seventy-two participants were trained to identify threats in pieces of luggage. Thereafter they were transferred to novel stimuli. Results can be summarized in support for the diversity of training hypothesis for preparing for novelty: To the best training for novel luggage screening situations is achieved using fewer items in a variety of categories. Copyright # 2010 John Wiley & Sons, Ltd. Target detection and decision-making tasks are pervasive. They are as common as finding flaws as a quality inspector, and as important and relevant for our health and society, such as a physician identifying a tumour on an X-ray image, a soldier determining the presence of a combatant in unfamiliar terrain, and an airport security officer looking for threats in passenger luggage. The terrorist attacks of 11 September changed the way security is addressed in American airports. However, much of the threat detection in luggage screening is still done by visual inspection rather than by automated methods. This is partly due to the complexity of visual images, the uncertainty and variability of what constitutes a threat, and the intricacies of the human decision-making process. Research that improves human accuracy of detecting potential threats and optimizes detection time has become a priority. Our applied research goal in the airport security context is to find ways to transfer skills acquired during training to the accurate detection of unfamiliar, novel targets. This implies the need to prepare security officers to identify novel items of known categories. For instance, they should be able to detect not only images of guns or knives they have encountered during training but also novel images of guns and knives that they have not yet seen before. We will refer to this as exemplar diversity in the following. As we describe below, there exists some evidence that humans can indeed learn to detect novel items of familiar categories. More challenging, security officers have to prepare for another kind of novelty that we call category diversity. They have to detect not only novel items of familiar categories, but also novel items of categories that by definition cannot be practiced during training and that will potentially not look like any weapons encountered in previous training. However, we would not expect luggage screeners to be prepared to detect any novel object but a novel exemplar of a novel category within a meta-category such as weapons or threats. For example, the meta-category of ‘cutting instruments’ that are not allowed in an aircraft includes knives, box cutters, machetes, etc. For this category, luggage screeners are likely to be trained on a subset of objects. This subset of objects should help screeners to detect other members of the metacategory of ‘cutting instruments’. Similarly, we would expect that when screeners are trained in multiple categories of weapons or threats, they would be able to detect other members of a more generic meta-category of ‘threats’. In this research we investigated how to best train to detect this kind of novel items. To do so, we focused on category diversity and exemplar diversity of stimuli presented during training in a luggage screening task to facilitate detection of novel exemplars of a novel category. There exist several research areas relevant to the goal of preparing luggage screeners for detecting novel items, but none of those provide sufficient and theoretically sound advice to solve this problem. In the following, we will briefly review the literature on skill acquisition, especially for perceptual learning, transfer of skills and category learning, and the implications for preparing for exemplar and category diversity.