Research

My neuroscience research is focused on understanding neural computations and information flow in brain networks.

What are neural computations?

The fundamental unit of information in the cortex is a pattern of activity across many neurons. These activity patterns propagate through neural circuits, and are changed by the circuits. These changes are neural computations. My long-term goal is to understand the computations in brains and understand how these computations create perception, decision, and action.

How do we approach these goals?

We develop cutting-edge experimental tools like two-photon holographic stimulation to analyze networks and population activity. Causal interventions are critical to understanding highly interconnected networks.

We collaborate with theoretical groups and use large-scale network simulations to enable close experiment-theory interactions that drive the science forward.

We are inspired by the architecture and computations of machine learning and artificial intelligence (ML, AI) systems. AI systems also perform computations at a network level. Is this process similar to the brain? As “interpretable AI” approaches seek to understand AI computations, can we use our two-photon optogenetic approaches to understand the brain?

Tools & Scientific Approaches

Current Research Themes

How do cortical networks process dynamic input?

We have found that the cortical recurrent network performs sequence filtering. 

How does learning change networks and computation?

We are examining plasticity in recurrent brain networks using cellular-resolution stimulation.

Beyond the cortex, how do other neural architectures in other brain areas enable the function of those areas?

We are using endoscopic imaging and stimulation to study this.

How do multiple brain areas interact to represent information and control information?

We are examining brain area interactions using mesoscopic two-photon imaging, behavioral assays, and two-photon perturbations.

Research Highlight: Inhibitory-Stabilized Networks

Excitatory and Inhibitory cells diagram

Cortical networks are strongly coupled: 

  • The excitatory-excitatory recurrent connections are strong in many or all cortical areas
    [Sanzeni & Akitake et al., 2021, Histed & Sanzeni, 2020].

  • Strong excitatory recurrent connections imply the cortex is well-described by inhibition-stabilized network (ISN) models. Inhibition tracks and balances excitation.

What this means in the brain:

  • The cortex operates in a balanced state where neurons receive large amounts of input.

  • Networks of realistic (conductance-based) require several scaling rules — of synaptic speed and also synapse number — to create balanced-state network operation. [Sanzeni, Histed & Brunel, 2022].

  • Balanced state/inhibition-stabilized operation of cortical operation renders excitatory cells sensitive to both increments and decrements of input. 

  • Suppression can result from withdrawal of input from other excitatory cells with little or no change in inhibitory firing.

Some of the approaches we use to gain insight into recurrent networks and neural computation include:

Foundational Research

My current research builds on discoveries from my prior work in: