Research
Motor Skill learning:
Dissecting complex learning processes
We are awed by Venus Williams’, Michael Jordan’s or Megan Rapinoe’s mastery of even the most complex skills. However, while maybe not as spectacular, is it not just as astonishing that each and every one of us can simply learn skill after skill? From tying our shoes to playing the piano to serving a beach volleyball – being able to learn complex motor skills is one of our biggest feats.
Decades of meticulous research have shown that learning comes in various flavors and have uncovered many mechanisms and neuronal circuits underlying different forms of learning. However, when it comes to more ‘complex’ learning processes, like motor skill learning, much work is still to be done. What really are the mechanisms allowing us to go beyond learning ‘simple’ associations and to learn how to generate countless flexible and intricate new behaviors – like new motor skills?
Key to our approach to these questions is a constant back-and-forth and mutual inspiration between the behavioral and neuronal levels. Our focus both on behavior and how it changes during learning and on the underlying brain-wide neuronal networks allows us to relate these levels and to build models of complex learning processes. Using this approach, we recently started to describe the intricate interplay between the basal ganglia, motor cortex and thalamus in motor skill learning and execution (1, 2).
The Wolff Lab aims to use and refine this approach to tackle the many facets of motor skill learning on a behavioral and neuronal level. As part of a community spanning across neuroscience, engineering and machine learning, we hope to not only help unravel the mysteries of motor skill learning, but of learning and memory in general.
Some of our Questions…
How do motor memories form and develop over the course of learning?
How do networks and mechanisms differ between learning processes?
Which additional nodes contribute to the distributed motor network?
How are memories for multiple skills stored in the same networks?
… and how we approach them
High-throughput behavioral training and analysis
Dissecting complex learning processes on a behavioral and neuronal level requires establishing clever behavioral paradigms, exploring critical behavioral variables and probing many nodes in the underlying distributed neuronal network. Since this asks for large numbers of time-consuming experiments, we will take advantage of a fully-automated high-throughput behavioral training system (3), which allows to train dozens of animals in parallel. Using modern machine learning techniques, we will precisely track the animals’ behavior throughout training and perform in-depth analyses beyond classical performance measures.
Tracking neuronal activity throughout learning
Observing how neuronal activity patterns change over the entire course of training is key in probing learning and memory formation. Furthermore, processes outside dedicated training sessions, e.g. during sleep, critically contribute to learning. To address both challenges, we will perform continuous (24/7) long-term electrophysiological recordings (4) over days, weeks and months. In combination with our automated analysis pipeline, this will allow us to precisely track learning-related changes in individual neurons and populations.
Monitoring and manipulating distributed learning networks
To explore how individual nodes - brain areas, circuits and defined neuronal subtypes - function and interact in the distributed networks underlying learning processes, we will extend our recordings to multiple brain areas, and combine them with a wide variety of acute and chronic neuronal activity manipulations (5). This will allow us to determine how activity manipulations affect both the animals’ behavior and the activity of identified circuit elements throughout the distributed networks (6,7).
Building frameworks and models for complex learning processes
We will use our large behavioral, physiological and functional datasets to build models of learning processes at different levels. We will move from simple conceptual models of learning, memory formation and storage to increasingly complex and powerful computational models, capturing our observations and making new testable predictions. We aim and hope for interactions and mutual inspiration with computational neuroscientists and machine learning researchers.