Dudman Lab — Research

Our Research

Learning at the speed of the animal mind

Animals are astonishingly fast learners. A naive mouse can home in on a hidden target in a handful of trials, learn which of six scattered options is the most valuable within minutes, covertly learn to avoid a future aversive event from seeing a few examples, or master a new task from a few large rewards — far faster than current models of reinforcment learning used to train today's large models. What are we missing?

We want to understand how the brain turns a handful of experience into robust, reliable knowledge. We study the neural circuits and algorithms behind rapid, flexible behavior: how the basal ganglia and cortex work conjointly to learn flexible, skillful action, how the structure of behavior itself accelerates learning, and how dopamine acts not simply as a reward signal but as an adaptive control knob that tunes how quickly we learn and how deeply we engage with the world.

Learning purposive, flexible actions

Although we often focus on the ability of a skilled movement to be executed rapidly and with precisely replicated kinematics, motor skill is also characterized by the flexibility with which an action can be executed while reliably achieving a goal. The ability to flexibly act with a range of vigor — amplitude, speed — is an essential aspect of skill.

Several lines of evidence suggest that basal ganglia play a critical role in the purposive control of movement vigor in mammals. To study the circuit mechanisms underlying the control of movement vigor we developed behavioral paradigms to study precise movement kinematics in mice performing voluntary movements of a joystick.

In a genetic model of Parkinson's disease we found that parkinsonian mice exhibit a profound enervation of joystick movements. Subsequent work showed that dopamine-dependent plasticity allows specific cell types in basal ganglia to exert a bidirectional, opponent control over movement vigor.

Neural Dynamics

Flexible action emerges from the coordinated activity of neural populations. We combine large-scale recordings, quantitative behavior, and circuit perturbations to ask how these patterns shape movement and change with learning.

Our work reveals how cortex and striatum jointly specify the details of an action, how population dynamics are shared across individuals and species, and how cortex can learn predictive control of innate movements from subcortical demonstration.

Models of Learning

How does a brain decide what is worth learning — and how fast? For decades, dopamine was cast as a teaching signal that reports errors in reward prediction. We found the story is richer, and more powerful: as animals learn directly from their own actions, mesolimbic dopamine sets the rate of that learning — an adaptive learning rate rather than a signed error signal. That reframing anchors the questions we pursue today.

Building on this, we recently found that dopamine in the ventral striatum can act as an adaptive, global learning rate — letting mice learn to exploit the best of six options scattered across a two-meter arena within minutes. And we discovered that reward magnitude, long minimized in the laboratory, is itself a lever on cognition: a few outsized rewards drive sustained dopamine release that makes animals learn more efficiently and stay engaged — and because engagement is the largest difference between individuals, bigger rewards markedly narrow the gap between fast and slow learners.

By pairing quantitative behavior with computational models and causal, cell-type-specific manipulations, we aim to explain not just what animals learn, but how the brain's own algorithms make learning so remarkably fast — even composing efficient trajectories toward a goal from only a handful of experiences.

Tools + Neuroscience

The work in our lab combines large-scale extracellular electrophysiology, optical recording, and intracellular electrophysiology with techniques to identify and perturb the activity of specific cell types and circuits in the mouse brain. This constellation of technical approaches requires significant development of hardware and software for data acquisition and analysis.

We are also actively involved in developing and refining molecular tools that allow us to identify, target, and manipulate specific cell-types in concert with physiology. In our recent work we have helped to develop and validate cell-type specific pharmacology in vivo using DARTs and developed a new class of optogenetic inhibitor exemplified by FLInChR.

We draw heavily upon the expertise of many staff scientists and engineers at Janelia. Some fruits of our labor can be found on the resources section where there are details about how to recreate hardware and download software.

Projects for Students

We use simultaneous, large scale neural recordings from multiple forebrain areas critical for descending control of novel motor skills. Optogenetic techniques for identifying and perturbing distinct cell types allows us to detail the logic of how distinct cell types implement distinct aspects of descending control.

In our recent work we have begun to examine how the activity of midbrain dopamine neurons evolve as a naive animal first learns about a reward predictive cue. Combining these new data with computational models provided the unique insight that dopamine neurons implement an adaptive learning rate that can improve learning.

Over the past several summers undergraduates in the lab have piloted diverse tasks for studying how goal-directed motor skills are executed and how flexible actions adapt to changing demands. Often such projects involve developing new hardware in collaboration with Janelia Experimental Technology.

Research