NINDS K99 Postdoctoral Fellow
Laboratory of Neural Systems
The Rockefeller University
New York, USA
ltian@rockefeller.edu
How are we so good at solving new problems, understanding new situations, and imagining new concepts such as a made-up animal? These involve the fundamental ability to construct new ideas by reusing known parts (sometimes called compositionality). My approach is to explain this ability in terms of both cognitive algorithms and how these algorithms are implemented in neural activity in the brain. I tackle this using approaches at the intersection of systems neuroscience, computational neuroscience, and cognitive science.
I record the spiking activity of many (100s of) neurons in animals performing difficult cognitive tasks requiring them to construct new solutions, or generalize. I analyze neural and behavioral data using various techniques, including computational modeling.
In animals who I trained to perforrm a drawing-like task requiring them to construct novel compositional action sequences to copy geometric figures (resembling handwriting), I and my collaborators discovered that primate frontal cortex has a representation of symbolically structured actions (e.g., a "circle" stroke abstraction) localized to neural populations in an an area called ventral premotor cortex (Tian et al., Nature 2026; COSYNE talk).
Evidence suggests that these action symbols correspond to motor concepts that allow us to internally construct and organize complex action sequences (e.g., to cook dinner), use tools, play music, learn sports, and otherwise interact intelligently with the physical world. Behavioral (Tian et al. NeurIPS 2020; NeurIPS talk) and brain imaging studies suggest this representation also exists in humans, and that it may even play a role in non-motor cognitive abilities such as imagining actions. Finding this representation in animals gives us a chance to eventually develop a neural mechanistic explanation.
In addition to representing units of action, the brain must also have mechanisms for recombining such units into new sequences. We found that rule systems for generating sequences (action grammars) are preferentially encoded by neurons in an area of the brain called presupplementary motor area (preSMA) (Tian et al., bioRxiv 2026).
I am currently a postdoc with Winrich Freiwald (Rockefeller University). I also work closely with Xiao-Jing Wang (NYU), and Josh Tenenbaum (MIT).
During my Ph.D. in neurophysiology and behavior at UCSF with Michael Brainard, I studied how the brain allows us to learn and executive fine motor skills; specifically, how circuits balance the need to reuse individual actions across sequential contexts (generalization) with the need for appropriate adjustments to actions within contexts (specialization) (Tian & Brainard, Neuron 2017; Veit, Tian, et al., eLife 2021; Tian et al., eLife 2023). In the songbird, I discovered that this balance is achieved by a hierarchical circuit in which one area encodes reused discrete actions, while a second cortical-basal ganglia area provides biasing signals to drive context-specific adjustments to these actions. This revealed that motor skill depends on a specific two-area hierarchical circuit mechanism.
Reach out if you'd like to chat! Or find me at one of these upcoming conferences:
(Updated Jan 9, 2026)