The UW Computational Neuroscience Center and the Shanahan Fellowship program have an exciting paid opportunity for pre-doctoral researchers at the interface of data and neuroscience. These roles are open to both current undergraduate and post-baccalaureate students.

Student fellows will be mentored by a current Shanahan Fellow, and they will gain hands-on research experience in neural computation, neural networks, and computational modeling/method development. Students will join a vibrant interdisciplinary research community with the opportunity to work with researchers at all levels at University of Washington and Allen Institute for Brain Science. As appropriate, fellows will be supported in opportunities to present and publish their work.

Applications are open now, with priority applications due July 18. Positions are open until filled.

Dr. Shuchen Wu

Bio: Shuchen Wu joined the Allen Institute and the University of Washington as a Shanahan Fellow in February 2025. She received her undergraduate degree in Physics and Applied Math from University of Rochester and a Masters degree in Neural Systems and Computation at the University and ETH Zurich. Shuchen received her Ph.D. in Theoretical Neuroscience from Max Planck Institute for Biological Cybernetics in Tuebingen, Germany, where she worked with Eric Schulz and Peter Dayan. During her PhD, she conducted experiments, developed theories and cognitive models on how humans learn patterns and entities from perceptual sequences. She later conducted a research visit at the explainable machine learning institute in Helmholtz Munich where she developed methods of interpreting neural population activities leveraging the cognitive principles of chunking. As a Shanahan Fellow, Shuchen is working on the question of how neural population activity adapts to regularities in the environment and generates a hierarchically structured behavioral repertoire.

Project:

How a complex, distributed computing system gives rise to a repertoire of behavior, and how the neural system adapts to the regularities and the structure of the environment, has been the key question concerning neuroscience and interpretable machine learning alike. While both have many computing units, discernable differences exist between the biological neural computation, such as a mouse brain, and the artificial neural computation, such as LLMs. This project explores interpretability tools in machine learning to discern computational patterns of neural circuits on a population level. It explores existing interpretability tools to find patterns of population-level activities in the mammalian cortex, utilizing rich datasets collected by the Allen institute such as the visual coding and visual behavior dataset. This project is aimed at relating patterns in neural population activity to behavior. It is suitable for candidates with a background in machine learning, cognitive science, data science, or neuroscience, and everyone in between.

The focus of the projects will be tailored to the interests of the candidate. A data science-focused project can be to improve on existing ways of extracting stereotypical population level activity, utilizing the current interpretability methods in ML and adapt them to biological data. A behavior-focused project can be to connect stereotypical neural activities with the annotated behavioral repertoire in animals or artificial neural networks, and to relate the occurrence of neural subpopulation abstraction with the context-dependent behavior in animals or artificial neural networks. A neuroscience-focused project is to study the relation between cell-types and the extracted patterns of population activities based on the Allen Brain Cell Atlas.

Preferred skills: Proficiency in Python. i.e, comfortable working with datasets using pandas, NumPy, and performing data visualization with libraries such as Matplotlib and Seaborn. Understanding of object-oriented programming principles in Python. Background in computer science, cognitive science, data science, or neuroscience.

Dr. Tim Kim

Bio: Tim Kim joined the Allen Institute and the University of Washington as a Shanahan Foundation Fellow in December 2024. He received his PhD in Neuroscience from Princeton University, where he worked with Carlos Brody and Jonathan Pillow. During his PhD, Tim developed unsupervised methods for discovering interpretable latent dynamics in high-dimensional neural data. Before that, he completed his undergraduate studies at the University of Pennsylvania, where he worked with Joshua Gold. As a Shanahan Fellow, Tim is analyzing new neural population datasets at the Allen Institute, and developing data-driven methods to bridge different levels of description, from individual cell types to interactions between multiple brain regions and behavior.

Project: How do large populations of neurons work together to solve complex tasks such as decision-making or motor control? One central premise in neuroscience is that the brain’s computations for performing such tasks can be succinctly represented as differential equations describing how population activity evolves over time. However, these equations, and the way in which they are biologically implemented, are often unknown.

This project will focus on developing a computational framework that can infer both the governing differential equations and their biologically plausible implementations, directly from large-scale neural recordings available at the Allen Institute. Our approach will combine the flexibility of modern machine learning (e.g., transformers) with the interpretability of classical computational neuroscience models.

This position offers close mentorship in both computational neuroscience and machine learning. The research assistant will receive hands-on training in developing interpretable models of neural activity. In addition to technical skill-building, there will be opportunities to engage in scientific discussions with researchers at the Allen Institute and UW, contribute to collaborative research efforts, and gain exposure to academic research workflows. Guidance will also be provided on navigating academic research and preparing for graduate study.

Preferred skills/qualifications: Bachelor’s degree in a quantitative field (e.g., neuroscience, computer science, mathematics, physics, or data science); Proficiency in Python, with experience in deep learning frameworks such as PyTorch, Flax, or similar; Strong interest in machine learning methods for science; Prior experience with modeling dynamical systems or neural data is a plus, but not required

Dr. Maria Tikhanovskaya

Bio: Maria Tikhanovskaya joined the Allen Institute and the University of Washington as a Shanahan Fellow in January 2025. Maria earned her Ph.D. in Physics from Harvard University in 2024, where she worked with Professor Subir Sachdev on developing theoretical frameworks to better understand the complex phase diagram of high-temperature superconductors. During her Ph.D., she also worked at Google Research as a student researcher, applying large language models to problems in physics and quantum chemistry. As a Shanahan Fellow, Maria is leveraging her expertise in physics and modern AI to contribute to advances in computational neuroscience, with a focus on bio-realistic modeling of brain circuits.

Project: Bio-realistic modeling is a rapidly advancing field in neuroscience. The key aspect of this approach is systematic integration of multimodal data from various sources into comprehensive network models of brain circuits. Such models often incorporate cell type information inferred from transcriptomic data, defining diverse types of neurons as building blocks of the simulated network. Approaches such as electron microscopy-based connectomics provide detailed information about network connectivity. Large-scale in vivo neurophysiology provides crucial data about neural activity, enabling benchmarking of the simulation results as well as predictions or inferences about neural activity and the underlying circuit mechanisms. Furthermore, the optimization of free parameters in bio-realistic brain circuit models has improved significantly in recent years through the use of deep learning, increasing both efficiency and fidelity. Together, these components support the development of comprehensive models that can be directly compared to experimental data, inform experimental design, and guide neuroscience research toward answering fundamental questions about brain function.
This project will contribute to the development of a bio-realistic model simulating neural activity with cellular resolution across a whole vertebrate brain. This role is well-suited for candidates with a strong interest in neuroscience and data science; prior expertise in either area is not required. The position will involve working with Python, programming on GPUs, conducting literature reviews, and constructing a well-organized data and code infrastructure. A strong commitment to acquiring and applying these skills is essential. The specific focus and balance of responsibilities can be adapted to align with the selected candidate’s background and interests.
Preferred skills: Strong background in quantitative and computational science in areas such as biology, physics, or related fields. Proficiency in Python, including experience with working on datasets using pandas and NumPy, and performing data visualization with libraries such as Matplotlib. Familiarity with working in codebases and using version control with Git. Knowledge of machine learning or deep learning is a plus, but not required. A background in neuroscience is preferred, though not essential.

Dr. Denis Turcu

Bio: Denis Turcu, a Shanahan Fellow at the Allen Institute and the University of Washington since September 2024, earned his undergraduate degree in Physics and Mathematics from Harvard University and his Ph.D. in Theoretical Neuroscience from Columbia University, supervised by Prof. Larry Abbott. His undergraduate research focused on motion-correction methods in C. elegans calcium imaging (Samuel lab), and quantum computing error-correction (Aspuru-Guzik lab). Denis developed an end-to-end model of active electrosensation during his Ph.D., incorporating physics, electroreceptor, and artificial neural network models to propose a new hypothesis on neural computation in the weakly electric fish G. petersii. He continues to apply these models to study social behaviors in these fish. Additionally, his other projects demonstrated the binary classification capability of very sparse recurrent neural networks resembling neocortical circuits, and assessed spike-sorting limitations of probes mismatched to the geometry of neural tissue. As a Shanahan Fellow, Denis explores biologically plausible learning using switching synaptic plasticity rules and, separately, models of locomotion in drosophila, focusing on sensory feedback and brain-environment interactions through muscles, joints and limbs.

Project: Without doubt, neural circuits perform impressive computations that enable rich behaviors. Even more impressive is that these circuits find remarkable solutions despite vast internal and external constraints. For example, neuronal development dictates the connectivity between cells, cell types, and brain regions; dynamic ranges and types of sensory receptors narrow the amount of observations an individual can make; mechanical restrictions of joints and muscles limit interactions with the environment; and physical laws govern all these sensory experiences and interactions. Additionally, neural circuits generate natural behaviors highly efficiently, robustly, and flexibly.

With that, we are seeking a research assistant for one year to pursue a project inspired by the following questions:

1) How do neural circuits find such elegant and efficient solutions, adapt previous ones, and learn new ones, despite facing significant biological and physical constraints?

2) How do these constraints shape the solutions implemented by neural circuits?

3) How can we adapt computational principles extracted from neural solutions in the presence of complex constraints to broader scientific problems, even outside of Neuroscience?

A successful project will help shed light on the interplay between neural circuit solutions to behavioral problems and the constraints that must be overcome. The specific project will be tailored to the interests of the candidate, and will ideally contribute to one of the following research directions:

a) Biologically plausible learning using switching synaptic plasticity rules

· This is a computational project that explores biological learning based on experimental observations that cortical synapse strengths are well modeled by a

binary state variable. This state variable hints that weak and strong synapses may be governed by different plasticity rules, and the goal of the project is to investigate the benefits and drawbacks of such a circuit. This project involves designing and training custom recurrent neural networks, and theoretical intuition about linear and non-linear dynamical systems.

b) Role of sensory feedback, and of connectome and bio-mechanics constraints in shaping neural computations that enable flexible and robust locomotion

· This project brings together experimental and simulated data, physics simulation engine, and deep reinforcement learning to model drosophila (fruit fly) locomotion. Currently available datasets provide great biological detail that constrain the models to reach more realistic solutions on locomotor tasks. This project involves many different modules that need to be combined to start diving into the role of the sensory feedback and the neural circuits that process this feedback during locomotion.

c) Neural mechanisms of social interactions, both collaborative and aggressive

· This is a collaborative project with an experimental lab focused on social interactions between freely behaving animals, the weakly electric fish G. petersii. It involves working with multi-modal experimental data, physics simulations and models, data analysis, curation and inspection, and generating testable hypotheses for near future experiments.

Preferred skills: Proficiency in Python, experience working with recorded or simulated data e.g. NumPy, SciPy, pandas, etc.; experience with standard data visualization libraries, e.g. Matplotlib, seaborn etc.; familiarity with deep learning libraries, e.g. PyTorch, JAX, TensorFlow, etc.; Coursework in computer science, machine learning, math, physics, statistics, computational neuroscience, or related topics; Interest in biologically oriented research questions focused on neuroscience, computational neuroscience, or related topics