The Action, Computation, & Thinking (ACT) Lab is directed by Samuel McDougle at Yale's Department of Psychology. The goal of our research is to leverage behavioral, computational, and neuroscience approaches to better understand the nature of human skill learning. The lab will be officially opening in July, 2020.
One of the defining characteristics of the human species is our massive repertoire of skills, and our astoundingly flexible capacity to learn new ones throughout life. From masters (Serena Williams, Mozart) to amateurs (the rest of us), learning a skill requires a mix of high-level knowledge, attention, and repetitive practice.
In this lab we are interested in studying how humans learn and perform skilled behaviors. Skilled behavior is complex, and skill acquisition relies on a variety of psychological processes and learning algorithms.
Here are some broad questions we are currently scratching our heads over:
• How do learned behaviors go from controlled and effortful to automatic?
• How do cognitive strategies and abstract thought facilitate motor learning?
• How does the brain represent goals, actions, and the rewards that reinforce our actions? How is this accomplished computationally?
• Which aspects of skill learning are explicit, and which are implicit? Do these processes interact?
• How do different systems (e.g., working memory versus reinforcement learning) vie for control during action selection?
• What is the role of the cerebellum in the cognitive aspects of learning?
We leverage multiple methodological approaches for investigating the psychology and neuroscience of human learning and memory.
The careful study of behavior is key to addressing psychological and neuroscientific questions. We use a variety of behavioral paradigms to study different aspects of skill learning, including sensorimotor learning (e.g., force-field learning, visuomotor adaptation, etc.), reinforcement learning (e.g., bandit tasks, probabilistic RL, stimulus-response map learning, etc.), and more traditional psychophysics (e.g., evidence accumulation). Our goal is to use behavioral experiments to test novel computational theories of learning.
Computational modeling has many functions for our research, from generating precise behavioral and neural predictions, to simply organizing our thoughts. Our modeling approach relies on various machine learning techniques (e.g., state-space modeling, Q-learning, simple neural networks) and cognitive psychological approaches (e.g., drift-diffusion models).
Neuroimaging and Neuropsychology
Functional Magnetic Resonance Imaging has made great progress as a neuroscience technique over the last two decades. We use a combination of model-based (e.g., McDougle et al., 2019) and multivariate fMRI methods to characterize teaching signals and map neural circuits underlying learning. We also work with populations with particular neural pathologies, such as spino-cerebellar ataxia, to study how different neural regions (e.g., the cerebellum, basal ganglia, etc.) contribute to different aspects of learning, memory, and movement. We use neuropsychology both to address basic research questions, and to try and develop improved neurorehabilitation protocols.