Exploring human rhythm through artificial neural nets
How does a population of neurons make predictions based on rhythms? How does a brain learn to synchronize the body with rhythms? And why are we inclined to move to rhythm? We are using three complementary neural network approaches to explore these questions: continuous-time recurrent neural networks (CTRNNs) that learn to anticipate rhythmic events, reinforcement learning (RL) that learns to synchronize through rewards, and predictive coding networks to explore how experiences pairing sound and movement during development can lead to spontaneous movement to sound. Each approach is evaluated against psychophysical data and neurophysiological recordings, providing multiple perspectives on how temporal processing emerges in neural systems. By demonstrating how rhythmic abilities can emerge from general learning processes, this research challenges the assumption that specialized neural timing mechanisms are necessary for rhythm perception and production, aiming to advance our understanding of temporal processing in the brain while providing insights into the development of rhythmic abilities in humans.
Rhythm as predictive processing
"Predictive Processing," prominent theory of the brain, posits that perceptual processes and action proceed according to a principle of continuously adjusting beliefs and our bodies (respectively) to minimize "prediction error," misalignment between what we expect and what we experience. We find that Predictive Processing offers an exciting lens onto how we experience and produce rhythm, and that our faculty of rhythm offers a domain in which we can further develop the art of mapping human experience onto Predictive Processing models. When our models are presented with rhythm, they describe the real-time adjustment of internal representations of beat phase, tempo, and metrical structure, and predict our motivation to move rhythmically, all based on a policy of prediction error minimization. We are now developing approaches to test and validate these models, which (when experimentally validated) will provide a language by which to describe connections between rhythm perception, movement, and cognition.
Subcortical structures in timing and rhythm
The cerebellum is a key brain region for predicting sensory consequences of our actions. Previous results suggesting that basic rhythm tasks do not depend on cerebellum, but if perceiving a beat is like performing an action as some researchers suggest, we expect that the cerebellum does get involved in predicting sounds in a rhythm. We will explore this through experiments combining rhythm and cerebellum-dependent classical conditioning of the blink reflex, and will model our results in the Predictive Processing paradigm described above, clarifying a role of cerebellum in rhythm, action, and cognition.
The basal ganglia region of the human brain and its dopaminergic inputs are thought to be important for rhythm perception and production, as evidenced by rhythm deficits when they are compromised by Parkinson's Disease. But it is as yet unclear how to understand the role of basal ganglia in rhythm. We have extended a model of action selection in the basal ganglia to describe how dopamine-mediated cortical-basal-ganglia loops can "lock in" a specific tempo for steady tapping or walking and stabilize it against noise. When dopamine is reduced, our model reproduces several Parkinsonian impairments, including tempo drift during rhythm production and freezing. This model makes predictions about the relationship between dopamine and tempo stability that we hope to test in future experimental collaborations.
Involuntary synchronization
It can be difficult for people to resist the urge to synchronize when tapping along to a rhythm. We are curious to investigate the factors that determine whether we synchronize with the rhythmic sounds we’re hearing, including attention, cognitive load, sensory motor feedback, and previous associations between the sounds and our own movements. The findings from these experiments will help inform conceptual and mathematical models of movement to rhythm.