Research Scientist/Engineer 3

University of Washington•Seattle, WA

21h

About The Position

The Division of Metabolism, Endocrinology and Nutrition Administration has an outstanding opportunity for a Research Scientist/Engineer 3 to support the Bowen laboratory. Field of research this position is engaged in: The Bowen lab leverages wide-scale neural recordings, predictive modeling, and continuous glucose monitoring with the goal of building foundational integrated (“multi-modal”) neural + behavioral disease-state models. The purpose of the research project(s) this position supports: The purpose of the research supported by this position is to develop a computational pipeline for transforming large-scale mouse behavior video data into standardized (canonical) representations that can be integrated with neural and physiological measurements. This project builds on existing methods, including a foundational model for mapping 2D mouse images to a canonical 3D mesh, with the goal of extending to full 3D reconstruction. This will enable inferring canonical 3D postures from single-view video data, which will provide a unifying modality for combining neural and physiological data across various experimental conditions. Research Sponsors/Stakeholders: HHMI Hanna Gray Fellowship

Requirements

Bachelor’s degree in Computer Science, Electrical Engineering, Applied Mathematics, Statistics, or a related field and four years of relevant experience in Machine Learning/Computer Vision/software engineering.
Equivalent education and/or experience may substitute for minimum qualifications except when there are legal requirements, such as a license, certification, and/or registration.
Demonstrated experience developing and/or implementing machine learning models in Python with PyTorch (or equivalent deep learning framework).
Experience with computer vision workflows (e.g., pose estimation, segmentation, tracking, multi-view geometry, 3D vision).
Strong software engineering fundamentals: version control, code review, testing, documentation, and reproducible experiments.

Nice To Haves

Ability to communicate technical concepts clearly and collaborate in an interdisciplinary research environment.
Experience with 3D reconstruction pipelines and/or mesh-based representations.
Experience scaling model training/inference on GPUs; familiarity with CUDA-adjacent performance considerations, distributed training, or workflow orchestration.

Responsibilities

Develop and validate canonical 3D posture estimation frameworks using core multi-view, ground-truth datasets, including methods for calibration, triangulation, and uncertainty quantification.
Design and implement a robust translation layer that generalizes these 3D posture models to large-scale 2D video datasets, enabling reliable inference across diverse recording conditions, camera configurations, and experimental paradigms.
Evaluate model performance across datasets and conditions, and iteratively refine algorithms to ensure scalability, reproducibility, and biological interpretability.
Build canonical behavioral embeddings that capture posture dynamics and action structure using modern representational learning approaches.
Apply temporal modeling techniques to learn multiscale behavioral sequences and transitions.
Establish a shared latent behavioral space that enables alignment and joint analysis of behavior with neural recordings and physiological signals.
Collaborate with domain experts to ensure learned representations are interpretable, biologically meaningful, and suitable for downstream scientific discovery.
Contribute to the building and maintenance of an end-to-end, production-grade pipeline encompassing scalable video preprocessing, model training, and inference workflows.
Implement GPU-accelerated training and inference, standardized evaluation protocols, and performance monitoring.
Manage versioned datasets and models, ensuring reproducibility, traceability, and compatibility with evolving methods.
Produce clear documentation, APIs, and interfaces to support downstream analysis modules and facilitate reuse by collaborators.
Serve as a primary technical point of contact for internal and external collaborators.
Advise collaborators on data formatting standards, behavioral annotation and sequencing, and integration pathways into the canonical pipeline.
Coordinate closely with lab members working on neural and physiological data integration, ensuring alignment between behavioral representations and experimental constraints.
Contribute to strategic planning by identifying technical risks, dependencies, and opportunities for methodological innovation.