|
| Aspects of Learning |
| Monday, May 28, 2012 |
| 9:00 AM–10:20 AM |
| 605 (Convention Center) |
| Area: TPC |
| Chair: Timothy C. Fuller (University of Nevada, Reno) |
| |
The Learn Unit: A Revised Conceptualization |
| Domain: Theory |
| BRANDON HERSCOVITCH (Simmons College), Ronald F. Allen (Simmons College), Amber Greenwood (ABA Consultation and Services, LLC), Alaina C Valentine (ABA Consultation and Services) |
| |
| Abstract: A revised conceptualization of the learn unit is presented. A discrete trial is a brief unit of instruction, and is generally conceptualized as having five essential components: the (1) cue, (2) prompt, (3) response, (4) consequence, and (5) inter-trial interval (Smith, 2001, p. 86). However, each of these broad components may actually contain multiple independent variables. For example, within the context of a matching pictures to objects program, the antecedent conditions programmed to evoke target responding include presentation of (1) the visual sample stimulus, (2) the visual comparison stimuli, (3) auditory sample (instructional cue), and (4) the programmed interval between the onset of the first antecedent condition and the offset of the last antecedent condition. For the purpose of procedural integrity, just scoring the antecedent as one variable may be insufficient to capture the accuracy with which the antecedent procedures, cumulatively, are implemented; the same goes for the other components. Implications of this conceptual analysis are significant for applied behavior analysis practitioners. |
| |
CANCELED: Learning Without Awareness: A Dissociation of Saying and Doing |
| Domain: Theory |
| FRANK HAMMONDS (Troy University) |
| |
| Abstract: Studies of learning without awareness involve an individual learning some experimental task and then responding to questions about what has been learned. If the individual does not adequately answer the questions, learning without awareness is said to have occurred. In this presentation, I will discuss the concept of learning without awareness. After introducing the topic, I will provide a brief review of the literature from the past several decades and outline my current research in the area. It is clear from the literature that learning without awareness does take place. However, I will show that awareness is nothing more than verbal behavior and that learning without awareness is simply behavior for which accompanying verbal behavior is lacking. Perhaps not surprisingly, verbal behavior is rarely explicitly mentioned in the learning without awareness literature. Behavior analysts obviously have a great deal to say about verbal behavior and thus about learning without awareness. Relevant statements from both behavior analysts and philosophers such as Gilbert Ryle will be included. |
| |
| Respondent Contributions Revisited: The Role Stimulus Correlation Plays in Derived Stimulus Relations |
| Domain: Theory |
| TIMOTHY C. FULLER (University of Nevada, Reno), Linda J. Parrott Hayes (University of Nevada, Reno) |
| |
| Abstract: A review of past and current trends in the area of respondent conditioning will be offered, paying particular attention to those areas where respondent interactions have been shown to be applicable. For example, the area of stimulus equivalence had received some attention from behavior analysts with regard to the role respondent processes may be responsible for the relations observed. The paper focuses on how inaccuracies in the description of respondent conditioning have led to a relative lack of attention to the contemporary advancements in respondent conditioning. A case is made that the analysis of derived relational responding is improved if respondent processes are taken into account. Additionally, it is the purpose of the paper to offer a conceptual analysis of derived relational responding, specifically with relations other than equivalence, wherein the extent to which respondent conditioning maybe applied. This is done by demonstrating how the experimental preparations found in the literature of derived relational responding are built from stimulus correlation rather then only operant processes as is stated by many contemporary researchers in this area. |
| |
| Value-Gradient Learning: Where Temporal Difference Learning and Operant Learning meet |
| Domain: Theory |
| EDUARDO ALONSO (City University London) |
| |
| Abstract: Temporal Difference Learning methods find optimal trajectories of behaviour assuming that Bellman's Optimality Condition applies. In the absence of a model, this has proved to be an impractical assumption as well as psychologically implausible.
We propose a new technique, Value-Gradient Learning (VGL), that works on Pontryagin's Minimum Principle. In so doing, convergence to optimal trajectories is guaranteed without local exploration. In control terms, the Hamiltonian of the system uses the gradient of the function to be minimized as a Lagrange multiplier on the constraints.
Now, this formulation fits behavior regulatory theories like Staddon's: The animals would learn to minimize the cost of diverting from the bliss point by balancing the constraints imposed by temporal and feedback functions against the behavior gradient (that is, against the derivatives of the cost function and the distribution of responses).
This new approach provides researchers with a mathematical tool to solve optimization problems in studying operant behavior; and it is psychologically well-founded. Besides, it helps relate Temporal Difference Learning to operant learning.
It is worth pointing out that VGL will find the solution, that is, the optimal distribution of responses for a given reinforcement schedule but, as a learning algorithm, it will also describe how organisms adapt to the unknown constraints in an optimal manner locally. |
| |
|
| |