UCLA

What conditions make learning efficient? Do adaptive methods — that use learner performance to arrange learning events — offer ways of improving learning? In this dissertation, I address these questions experimentally in human subjects across domains of factual and perceptual learning. Six experiments focus on methods of scheduling the order of presentation of items during a learning session with the goal of improving long-term retention. We introduce a method of determining spacing schedules using an adaptive, computer-based algorithm. Space between presentations is dynamically calculated as a function of a learner's response history and reaction time. This work is based on the spacing effect in memory: when learning a set of facts, scheduling time between repeated presentations or practice has been shown to improve learning. Larger gaps between presentations of an item contribute to increases in the strength of that item in memory. The primary experimental manipulation in these studies is a comparison between fixed schedules of practice and adaptive schedules of practice. Three crucial issues are tested. First, the effects of spacing in adaptive, computer-based schedules are compared to predetermined schedules that have fixed intervals of spacing. Second, adaptive schedules are compared to schedules that are completely randomized, and the difference between schedules with and without dropout is compared. Finally, the benefits of adaptive scheduling are assessed using perceptual category learning. Results of the first set of experiments showed that an adaptive scheduling algorithm produced greater learning gains than fixed schedules, when the total number of presentations was limited. These gains were measurable after a one-week delay. Further, when fixed condition schedules were closely matched to adaptive scheduling in overall item spacing characteristics, adaptive schedules still outperformed them in terms of learning gains at immediate and delayed tests. Results of the second set of experiments, where learning was allowed to continue until learners met learning criteria, showed adaptive scheduling comparing favorably with random presentation schedules. Adaptive schedules where items were dropped after learning criteria were met produced greater learning efficiency (learning gains per trials invested), better than random presentation schedules with or without dropout These efficiency gains were maintained at a delayed test. In a final set of experiments, adaptive schedules increased learning efficiency during perceptual category learning reliably more than random schedules as well as schedules that were adaptive but included some initial blocking or massing of category exemplars. Adaptive scheduling also improved the fluency gains of students learning 3D chemical structure in an introductory community college chemistry course. The results in this thesis relate to state-of-the-art computer adaptive methods of teaching, as well as contemporary models of learning, memory, perception, categorization and human performance. These studies contribute to research in learning and memory, are of broad interest to educators who are concerned about student learning, and inform attempts to connect models of cognition with technology-based tools.