Computational Neuroscientist, Modeling / Theory

Astera Institute•Emerybille, CA

18h

About The Position

Computational Neuroscientists at Astera Neuro develop and lead research programs aimed at understanding the representations and dynamics underlying conscious access, and at converting that understanding into the ability to steer the system. The work sits at the intersection of cognitive theory, large-scale neural data, and machine learning, and draws on recordings from our primate, rodent, and human programs. The opportunity is to create a new theoretical edifice for how the brain builds an internal model of the world, on data of a scale, breadth, and quality that has not previously existed. Computation sits at the core of Astera Neuro, and the role is built around tight coupling with experiment: models are expected to make testable predictions and to propose the next experiment rather than wait for it. Depending on your background and interest, you may focus on modeling large-scale neural data, extracting structure from recordings that span many areas, sessions, animals, and species, or on theory, abstracting the key computational principles out of the data and into new NeuroAI architectures. We welcome computational scientists trained outside neuroscience: backgrounds in control theory and robotics, theoretical physics, and statistics and machine learning are all highly valued here, and we expect that the theory we are after will need to draw on all of them. Title and scope are calibrated to track record.

Requirements

Conviction that the brain's internal model can be understood in full, and recognition that getting there requires a kind of science no single academic lab can do. We are betting on scale, on deep collaboration across science and engineering, and on open sharing of ideas and expertise in a full-stack environment.
Appetite for building theory rather than applying it. The framework that explains how a physical system converges on a stable internal model of the world does not yet exist. We are looking for people who want to build it, and who are willing to work on the data long enough to find out what it demands.
Strong computational skills, with fluency in Python and modern machine learning frameworks such as PyTorch or JAX, and comfort with large-scale data pipelines. You move between theory, data, and code fluently.
Comfort building on shared infrastructure rather than private projects. Recordings and perturbations will be registered onto a common whole-brain functional and anatomical atlas, and datasets are designed to stitch across sessions, animals, and paradigms. This means designing experiments others can build on, and building on theirs.
Experience with latent-variable and state-space models of neural population activity, such as GPFA, LFADS, switching state-space models, or recurrent network models fit to data.
Commitment to open science. We will release tools, data, and methods, and we aim to create a new dynamic of rapid, open exchange in neuroscience.
PhD in computational neuroscience, neuroscience, physics, statistics, applied mathematics, electrical engineering, computer science, control theory or robotics, or a related field, or equivalent research experience.
Graduate work or research experience in neuroscience is a plus but not required.
We value demonstrated skill and relevant experience above credentials.

Nice To Haves

Demonstrated ability to lead a computational or theoretical research project end to end, from question and formulation through implementation, analysis, and publication.
Depth in dynamical systems, including fixed-point and attractor analysis, stability and bifurcation structure, and the fitting of dynamical models to noisy, partially observed data.
Hands-on experience analyzing large-scale neural datasets from electrophysiology, two-photon imaging, or comparable methods, including high-density recordings such as Neuropixels and modern spike-sorting and quality-control pipelines such as Kilosort.
Background in control theory, optimal control, or robotics, particularly as applied to steering high-dimensional systems toward target states.
Background in statistics or theoretical physics, with a serious interest in applying it to neural systems.
Experience designing or implementing real-time or closed-loop decoders and model-guided stimulation.
Experience with foundation-model and self-supervised approaches applied to neural or behavioral data, and interest in the correspondence between artificial and biological representations.
Contributions to open-source neuroscience or scientific computing projects.
Ability to work at close quarters with experimental neuroscientists, software engineers, and hardware engineers in a fast-moving, multi-team environment.
Track record of owning a modeling or theory agenda end to end, including framing the question, collaborating with experimentalists, and delivering results (expected at the senior or principal level).
Experience making and defending modeling tradeoffs across interpretability, predictive accuracy, and experimental utility (expected at the senior or principal level).
Comfort working across the stack, from data infrastructure and analysis pipelines through to theory and machine learning (expected at the senior or principal level).
History of mentoring scientists or leading technical initiatives (expected at the principal level).

Responsibilities

Build models of compositional neural representation, grounded in cognitive theory and fit to recordings: how the brain binds content to variables, composes structured thought, and updates it.
Build coupled dynamical systems models of the interactions between multiple brain areas, and test them against simultaneous multi-area recordings.
Analyze the fixed-point structure of these systems and characterize the landscape of stable states underlying percepts, thoughts, and internal states.
Derive how to sculpt inputs, using our optogenetic and electrical stimulation technology, to drive the system to chosen stable points, and validate those derivations in closed-loop write-in experiments.
Stitch data together across subjects and species into foundation models of neural activity, registered onto a common whole-brain functional and anatomical atlas.
Propose and help design new experiments.
Identify the measurement or perturbation that would most sharply separate competing models, and work with the experimental teams to run it.
Abstract key computational principles from the data into new NeuroAI architectures, in some cases in direct collaboration with Astera AI.
Mentor research engineers, and at the senior or principal level, more junior scientists.
Contribute to hiring, onboarding, and lab culture.
Contribute to publications, talks, open data and tooling releases, and engagement with the broader scientific community.