Research

Executing successful movements requires the brain to predict the consequences of actions. It is believed that the brain builds internal models of our body and the environment in order to simulate the sensory and motor outcomes of movements. 

Due to the constant changes in our body and environment (for instance, those due to fatigue,  tool-use, or disease) these models require constant re-calibration, called motor adaptation, to keep us moving in predictable ways.

Where in the brain these models reside, how they are formed, and how they are updated following bodily or environmental changes remains unclear. 

The goal of the laboratory is to reverse engineer the neural circuits that drive adaptive motor behavior. We hope that by understanding the neural basis of adaptive motor control we can open new avenues in therapeutic research for neurological disease, help build better machine learning tools, and crucially, provide fundamental insights into brain function.  

Here are some questions that guide us:

• how are internal models represented in the neural code?

• what are the sensory and motor cortical contributions to motor adaptation?

• how are multiple areas across the brain efficiently sharing information during learning?

• how do (biological) neural networks enable lifelong learning?

behavior, models, & neural data

We believe behavior is an essential component to understanding neural function. As part of our quest to better understand behavior, we develop new tools to study more complex and natural movements. We develop tools, like DeepLabCut, to perform markerless pose estimation and behavioral analysis from any species in a multitude of settings. We also have developed a set of skilled motor tasks where mice can learn from a dynamically changing sensory landscape.

By combining concepts from machine learning and optimal motor control with the power of the mouse's genetics and accessibility, our lab aims to uncover fundamental principles that guide motor adaptation, learning, and motor control.

We are using the latest techniques in 2-photon and deep brain imaging (including utilizing multi-area imaging with a 2-photon mesoscope), to uncover the neural correlates of adaptive behavior. We use optogenetics and chemogenetics to test what roles diverse areas have during behavior. Furthermore, we develop new computational models and tools to generate testable hypotheses and analyze our data.

People

Collaborators:

Alumni:

Adrian Hoffmann
Masters student | University of Tübingen, Neural Information Processing.
hoffmann@rowland.harvard.edu; Adrian completed his master's thesis in the lab through the University of Tübingen in 2018. He received a BS in Physics from Heidelberg University. Adrian is now a PhD student in the group of Prof. Helmchen in Zurich.

Melody Tong - Undergraduate Researcher|Harvard College Class of '18
- Melody was co-mentored by Mackenzie and Nao Uchida. Her thesis work was focused on characterizing a rapidly learned freely-moving reaching & pulling task in mice.
She is now attending medical school at NYU.

Funding:

We gratefully acknowledge the funding sources that make our research possible:

Current:

2018 - 19 | Mind, Brain Behavior Faculty Award

2017 - '22 | Rowland Fellowship to M. Mathis

2018 | NVIDIA GPU Grant Award

Past:

2017 | WATB/Project ALS - Postdoctoral Fellowship to M. Mathis

2013 - '17 | NSF Graduate Research Fellowship to M. Mathis
 

Affiliations:

The Rowland Institute at Harvard
Mind Brain Behavior - Interfaculty Initiative
Harvard Brain Science Initiative

 

News

Our work on building the DeepLabCut toolbox, a deep learning method to perform markerless pose estimation was published in Nature Neuroscience, congrats to Alexander Mathis and co-authors!: rdcu.be/4Rep

Our first preprint from the lab, in collaboration with Matthias Bethge, is up on arxiv (check out our abstract page)

December 2016