Maayan Levy

NOMIS Fellow

Maayan Levy is a NOMIS Fellow at the Institute of Science and Technology (IST) Austria.

Levy received her bachelor’s degree in psychology and humanities (summa cum laude) from Tel-Hai Academic College (Israel) and proceeded to complete her master’s degree in biopsychology at the Hebrew University of Jerusalem. In 2012 she joined the machine vision startup Cortica, working as a software engineer for three years, but could not stay away from research. She earned her PhD in computational neuroscience from the University of Chicago in 2021. Her dissertation examined the network structure of pairwise statistical dependencies between the activity of neurons in the mouse primary visual cortex and the implications of this organization to the encoding of sensory information.

Synaptic connections in the brain undergo changes known as plasticity. Ample research has provided insights into the cellular and molecular underpinnings of plasticity in the single synapse resolution, whereas present-day machine-learning (ML) algorithms generate rapid learning of complex tasks on the network level. The challenge of bridging experiments and theory takes two specific, intertwined forms: First, experimental induction protocols for synaptic plasticity often use unrealistic inputs to neurons and produce mixed and frequently noisy results in vivo. It is often unclear what the optimal protocol would be to tease apart the complex contributions of multiple parameters to synaptic plasticity and in turn to activity and behavior. Artificial networks also typically employ connectivity parameters and learning rules that are not biologically plausible, rendering the interpretability of the results difficult. Second, it is unclear how plasticity as it is understood on the single synapse level is distributed and coordinated across a population of neurons, especially given the diversity of neuronal cell types.

As a NOMIS Fellow, Levy proposes a synergy between theory and experiments: First, she will develop machine learning approaches to learn the best protocols for probing the plasticity of neural systems. Induction protocols used in the Jonas group to study plasticity in the mossy fiber synapse of the hippocampus will be optimized with a framework inspired by previous work in Tim Vogels’ lab. This research project pioneers the use of ML methods to explore, refine and guide experimental protocols. It will pave the way for efficient method development in virtually any discipline in the biological sciences, alleviating difficulties associated with trial-and-error and replication. Levy will then construct neural network simulations to explore the minimum requirements for plasticity on the network level, asking, what are the quantity, magnitude and architecture of efficacy changes that are necessary for learning? This will result in the ability to compare learning in different topologies composed of varying cell types, or networks such as naïve and expert, generating hypotheses for anatomical, functional and developmental studies in the hippocampus. The project embraces the complexity and diversity of plasticity in real circuits, and the results of this work will enhance our understanding of the formation of short-term memories.

Feature image: Data-informed and biologically realistic neural network simulations are allowed a limited number and magnitude of local synaptic changes. Multiple simulations constitute particles in a swarm optimization, revealing the topological reorganization underlying memories.