Flexible problem-solving and reasoning

Primates can solve novel problems through logical and stepwise reasoning. No two real-world situations are the same, and how one ‘figures out’ a solution may be similarly variable. Studying reasoning has thus been challenging. How should one investigate the neural basis of hidden mental operations whose timing and nature are uncertain, and are unlikely to ever unfold the same way twice? 

At the heart of our lab’s mission is a bold scientific challenge: to uncover how the brain transforms goals and knowledge into flexible, intelligent behavior. This capacity for abstraction, deductive reasoning, and cognitive flexibility - which drive phenomena such as ‘zero-shot learning’ - is one of the most profound open questions in systems neuroscience and cognitive psychology. To meet this challenge, we are developing sophisticated behavioral tasks (e.g., the 'flexible deductive reasoning task' on the left) that enables us to interpret neural activity during unique, one-off cognitive events, when a subject sees, does, or figures out something for the very first time. 

Ultra-large-scale brain-wide electrophysiology 

To understand how flexible, intelligent behavior emerges from the brain, we need to study not just isolated neurons or localized circuits, but entire brain-wide networks working in concert. LOGIC is developing powerful new tools to do exactly that. As one recent example, we deployed numerous Neuropixels-NHP probes simultaneously to record thousands of neurons across multiple brain areas in real-time. These measurements have given us an unprecedented view of neural activity during cognitive behaviors on individual trials. This represents a huge leap forward. For perhaps the first time in history we’re beginning to study neural activity with the same richness and nuance that psychologists have brought to behavior, enabling us to interpret individual mental strategies and uncover the neural basis of higher-order logical reasoning. We are continuing to push the state-of-the-art in neurotechnology, developing new tools to both record and perturb brain-wide neural activity in real time on single trials - enabling truly insightful investigations of cognition as it unfolds. 

Computation through neural population dynamics 

At the heart of every research project in our lab is a fundamental question: What computational problem is the brain solving to achieve its current behavioral goal? While we begin with hypotheses at the language level, the investigation quickly becomes quantitative, focusing on mechanistic models that align with what a recurrent network of neurons can realistically implement. We are deeply interested in the behavior of individual neurons, but as we record from frontal brain areas, we realize that these computations are not carried out by isolated subsets of neurons. Instead, the neural responses we measure are a complex mixture of multiple overlapping computations, often occurring simultaneously. This requires a powerful framework capable of disentangling these mixed signals from individual neural responses.

One such framework is called computation-through-dynamics. Its core thesis is that, embedded within the responses of single neurons, there exists a smaller set of “computationally meaningful” signals. By identifying and studying these signals, we can form rigorous hypotheses about the underlying computational processes of the circuit. What’s particularly compelling is the consistency of this framework across a wide range of studies, brain areas, and model organisms in systems neuroscience. These “computationally meaningful” signals provide profound insights, not only in terms of what they represent but also in how they evolve over time (i.e., their dynamics, which perform the computation). As a result, nearly every project in our lab leverages this approach to investigate neural computation.