R. Sun, X. Zhang
2019
Citations
2
Influential Citations
4
Citations
Journal
Proceedings of the Twenty-Fourth Annual Conference of the Cognitive Science Society
Abstract
Top-Down versus Bottom-Up Learning in Skill Acquisition Ron Sun (rsun@cecs.missouri.edu) Xi Zhang (xzf73@mizzou.edu) Department of CECS, University of Missouri, Columbia, MO 65211, USA Abstract This paper studies the interaction between implicit and explicit processes in skill learning, in terms of top-down learning (that is, learning that goes from explicit to implicit knowledge) vs. bottom-up learn- ing (that is, learning that goes from implicit to ex- plicit knowled e). Instead of studying each type of knowledge emplicit or explicit) in isolation, we highlight the interaction between the two types of processes, especially in terms of one type giving rise to another. The work presents an integrated model of skill learning that takes into account both im- plicit and explicit processes and both top-down and bottom-up learning. We examine and simulate hu- man data in the Tower of Hanoi task. The paper shows how the quantitative data in this task may be captured using either top-down or bottom-up ap- proaches, although top-down learning is a more apt explanation of the human data currently available. The results demonstrate the difference between the two different directions of learning (top-down vs. bottom-up), and also provide a new perspective on skill learning in the Tower of Hanoi task. Introduction This paper studies the interaction between the im- plicit and explicit processes in skill learning. It explores two directions of skill learning: top-down learning and bottom-up learning. Top-down learn- ing goes from explicit knowledge to implicit knowl- edge, while bottom-up learning goes from implicit knowledge to explicit knowledge. Instead of study- ing each type of knowledge (implicit or explicit) in isolation, we want to highlight the interaction be- tween the two types of processes, especially in terms of one type giving rise to another. In this work, we want to test possibilities of bottom-up learning vs. top-down learning. We do so by using the task of Tower of Hanoi, which is arguably a typical benchmark problem in high-level cognitive skill acquisition and has been used in many previous studies of skill acquisition, cognitive mod- eling, and cognitive architectures (see, e.g., Proc- tor and Dutta 1995, Anderson 1993, Anderson and Lebiere 1998). To explore bottom-up and top-down learning, the work presents an integrated model of skill learning that takes into account both implicit and explicit processes and both top-down and bottom-up learn- ing, although the model was initially designed as a purely bottom-up learning model. We examine and simulate human data in the Tower of Hanoi task. The work shows how the quantitative data in this task may be captured using either top-down or bottom-up approaches, although we will show that top-down learning is a more apt explanation of the human data currently available in this task. Overall, the result of our simulations suggests that both directions are possible in human cognitive skill acquisition, and the actual direction may be either bottom-up or top-down (or a mix of both), depend- ing on task settings, instructions, and other vari- ables. These results demonstrate the two dilferent directions of learning (top-down vs. bottom-up), and also provide a new perspective on skill learning. Top-Down vs. Bottom-Up: The CLARION Model The role of implicit learning in skill acquisition and the distinction between implicit and explicit learn- ing have been widely recognized in recent years (see, e.g., Reber 1989, Stanley et al 1989, Willingham et al 1989, Anderson 1993, Seger 1994, Proctor and Dutta 1995, Stadler and Frensch 1998). However, although implicit learning has been actively inves- tigated, complex and multifaceted interaction be- tween the implicit and the explicit and the impor- tance of this interaction have not been universally recognized. To a large extent, such interaction has been downplayed or ignored, with only a few no- table exceptions (e.g., Mathews et al 1989, Sun et al 2001). Similar oversight is also evident in compu- tational simulation models of implicit learning (with few exceptions such as Cleeremans 1994 and Sun et al 2001). Despite the lack of studies of interaction, it has been gaining recognition that it is difficult, if not im- possible, to find a situation in which only one type of learning is engaged (Reber 1989, Seger 1994, Sun et al 2001). Our review of existing data has indicated that, while one can manipulate conditions to empha- size one or the other type, in most situations, both types of learning are involved, with varying amounts