|
| Innovative Analytical Approaches in Behavior Analysis: From Basic to Applied Research |
| Sunday, May 26, 2024 |
| 8:00 AM–9:50 AM |
| Convention Center, 200 Level, 203 AB |
| Area: PCH; Domain: Theory |
| Chair: Marc J. Lanovaz (Université de Montréal) |
| Discussant: Donald A. Hantula (Temple University) |
| CE Instructor: Marc J. Lanovaz, Ph.D. |
| Abstract: In the past decade, advances in data collection and processing have led scientists to examine the application of innovative analytical approaches to a wide variety of problems. These innovative approaches are also making their way across all domains of behavior analysis. The purpose of this symposium is to present some of these novel approaches as applied to data from nonhuman and human behavior. The first presentation examines how researchers may leverage technology to collect spatiotemporal data and test innovative hypotheses with rats. The second address compares different machine learning algorithms to identify schedules of reinforcement in pigeons. For the third presentation, researchers have conducted a study on applying genetic algorithms to produce artificial organisms, which can then be used to develop models to identify the reinforcement schedules and magnitude of automatically reinforced behavior. Finally, the symposium ends with a presentation on the application of disequilibrium theory in clinical settings. Altogether, the presentations provide exemplars of innovative data analyses that contribute to both basic and applied research in behavior analysis. |
| Instruction Level: Intermediate |
| Keyword(s): artificial intelligence, data analysis, machine learning, spatiotemporal data |
| Target Audience: Advanced graduate students, BCBAs, and BCBA-Ds with intermediate competencies in conducting data analysis on behavioral data |
| Learning Objectives: At the conclusion of the presentation, participants will be able to: (1) Describe at least one novel analytical approach applied to nonhuman animals, (2) describe at least one novel analytical approach applied to human behavior, and (3) explain how novel analytic approaches may contribute to the development of behavior analysis. |
| |
| Spatiotemporal Data and Machine Learning: Shaping the Future of Behavior Analysis and Reinforcement Schedules |
| ALEJANDRO LEON (Universidad Veracruzana), Varsovia Hernandez Eslava (Universidad Veracruzana) |
| Abstract: This presentation introduces an innovative approach to behavior analysis, breaking away from the conventional focus on single responses and their timing. Instead, we delve into the continuous spatial aspects of behavior, a vital component of natural environments. We employ cutting-edge computer vision techniques for real-time spatial behavior monitoring, shedding light on previously unexplored insights. Our research highlights how spatial dynamics, including variables like entropy, are profoundly affected by reinforcement schedules. We introduce novel reinforcement schedules derived from real-time spatiotemporal data, showcasing two instances of behavior analysis under individual and social conditions. These examples reveal how organisms respond to reinforcement schedules rooted in spatiotemporal factors. We harness machine learning tools to conduct multidimensional behavior analysis, considering variables like reinforcement rate, distance traveled, velocity, proximity to reinforcement sources, the time organisms spend in reinforcement zones, spatial anticipation of reinforcement, and entropy, among others. This presentation underscores the importance of uniting spatiotemporal data with behavior analysis and reinforcement schedules. By doing so, we expand the horizons of behavioral science, allowing it to encompass a broader range of natural settings and variables. |
| |
Predicting Animal Learning Histories With Artificial Intelligence (AI): Insights From Snapshot Data |
| ANNA PLESSAS (Auckland University of Technology), Josafath Espinosa-Ramos (Auckland University of Technology), Dave Parry (Murdoch University), Sarah Cowie (University of Auckland, New Zealand), Jason Landon (Auckland University of Technology) |
| Abstract: Past experiences shape behaviour and future actions are often based on current behavioural patterns. This research introduces an innovative tool to retrodict unknown learning histories from limited behavioural datasets extracted from binary choices made by laboratory pigeons. We extracted a snapshot of choice behaviour, including a 5-second post-reinforcer-delivery observation window. A Spiking Neural Network (SNN) was designed to generate retrodictions of learning histories. In Experiment 1, retrodictions were achieved, indicating that pigeons' binary choice snapshots contained sufficient information for the SNN model. Experiment 2 compared the SNN with other deep neural network models, with the SNN demonstrating superior retrodiction performance based on additional metrics. Experiment 3 illustrated the SNN's adaptability to novel, small-sized datasets of varying learning histories; personalised modelling further improved retrodictive performance. Experiment 4 validated the method's reliability through transfer learning techniques, highlighting the SNN's robust retrodiction capabilities. The SNN demonstrates proof of concept by discerning individual differences in learning patterns and the relationship between behaviour and learning history. The SNN's adaptability to small datasets makes it economical and easy to test in real-world conditions. Implications for both clinical and basic research will be discussed, including research in the use of SNN with human behaviour datasets. |
| |
| Detecting the Magnitude and Schedule of Automatic Reinforcement Using Artificial Neural Networks: A Simulation-Based Study |
| MARC J. LANOVAZ (Université de Montréal) |
| Abstract: Given that the experimenter cannot directly manipulate the reinforcer maintaining the behavior, identifying the magnitude and schedule of automatic reinforcement remains a challenge for researchers and practitioners. One potential solution to this challenge is to combine the temporal dimensions of behavior with machine learning to develop models that can detect magnitude and schedules. However, the lack of access to an external reinforcer maintaining the behavior makes it difficult to develop models using nonhuman and human behavior. To this end, we used the evolutionary theory of behavior dynamics to simulate 250 artificial organisms that engaged in automatically-reinforced behavior under two different RI schedules with low and high magnitudes of reinforcement. A hold-out cross-validation approach was then applied to train and test artificial neural networks with rate and IRT data. The results showed that machine learning could detect changes in reinforcement magnitude with a high level of accuracy. That said, machine leaning did not considerably improve the identification of schedules. Further work in this area of research should thus focus on identifying other features of single-case data that would improve the models, especially in discriminating schedules. |
| |
| Applying Disequilibrium Theory in Clinical Settings: Considerations for Practice and Research |
| HUNTER KING (Kennedy Krieger Institute, Johns Hopkins School of Medicine), John Falligant (Kennedy Krieger Institute/Johns Hopkins University School of Medicine) |
| Abstract: Derived from Timberlake and Allison’s (1974) response deprivation hypothesis, disequilibrium theory defines reinforcement and punishment in terms of instrumental and contingent activities (rather than stimuli). An instrumental activity is a response that produces an opportunity to engage in a contingent activity. Disequilibrium theory affords a novel technique for measuring behavior in applied settings as it focuses on the ratio of instrumental to contingent activities during free operant baselines to formalize the conditions under which a deprived organism will modify instrumental responding to access to the contingent activity. In recent years, researchers have evaluated the model’s ability to predict the desired effects of an intervention in human operant arrangements and clinical settings. The corpus of this literature supports the model’s ability to predict and control behavior, providing clinicians with a precise analytic method for determining contingency arrangements to address problems of social importance. This presentation provides an overview of disequilibrium theory, describes the process for how to collect data in a manner that facilitates analysis of behavioral events using this approach, and highlights several applied demonstrations of the model for making predictions about clinically significant behavior in treatment-resistant populations. |
|
| |