Data Scientist II - AI Assurance Researcher

Responsibilities

PNNL is seeking a Data Scientist II - AI Assurance Researcher who has experience in understanding, exploring, and manipulating the internal mechanics and behaviors of AI models and can provide valuable insights into the decision boundaries and mathematical fingerprints of data properties. The selected candidate should have extensive experience training models, accessing and working with data embeddings, the ability to derive theoretical queries from empirical results, and demonstrate the capacity to translate research papers and findings into mission relevant insights and tools.

Key Responsibilities:

• Design and execute rigorous machine learning experiments following community best practices, including large‑scale training and evaluation on HPC infrastructure.
• Evaluate AI systems across multiple dimensions, including standard performance metrics, generalization, robustness, and out‑of‑distribution behavior.
• Analyze the internal structures and representations of deep learning models, with emphasis on large language models and large vision models.
• Assess empirical results to identify trends, inform research direction, and recommend next steps.
• Write high‑quality, well‑documented research code that supports reproducibility and collaboration.
• Contribute to publications and technical reports that communicate findings to both internal and external audiences.
• Collaborate with senior researchers, engineers, and cross‑functional teams to integrate research into broader systems and workflows.
• Conduct work in secure environments with adherence to operational security requirements.

This position is based in either Richland, WA or Seattle, WA and requires an onsite presence.

• BS/BA and 2 years of relevant experience – or
• MS/MA – or
• PhD

• Degree in computer science, engineering, mathematics, or a related field.
• Familiarity with the current ML research landscape, particularly in explainable AI, adversarial machine learning, AI safety, and the science of deep learning.
• Hands‑on experience analyzing the internal structures of deep learning models, particularly large language models and large vision models.
• Proficiency in PyTorch and associated deep learning libraries.
• Experience designing and executing experiments on HPC systems.
• Experience in translating research prototypes into deployable tools and capabilities.
• Demonstrated strong communication and collaboration skills, with the ability to work effectively in a multi‑disciplinary team environment.

• U.S. Citizenship.
• Background investigation:
  Applicants will be subject to a Federal background investigation and must meet eligibility requirements for access to classified matter in accordance with 10 CFR 710, Appendix
  
  B.
• Drug testing:
  All Security Clearance positions are Testing Designated Positions, which means that the applicant selected for hire is subject to pre‑employment drug testing, and post‑employment random drug testing. Applicants must be able to demonstrate non‑use of illegal drugs, including marijuana, for the 12 consecutive months preceding completion of the requisite Questionnaire for National Security Positions (QNSP).

New employees must obtain and maintain a HSPD‑12 Personal Identity Verification (PIV) credential. The tier of federal background investigation is determined by job duties and national security or public trust responsibilities.

• Illegal drug activities, including use, supply, possession, or manufacture within the last 1 to 7 years, are declared.
• Foreign national candidates who have not resided in the U.S. for three consecutive years are not eligible for the PIV credential and will require a Local Site‑Specific Only (LSSO) determination. After meeting the three‑year residency requirement, a PIV credential will be required.

DOE prohibits employees and contractors from having any affiliation with the foreign government of a country DOE has identified as a “country of risk” without explicit approval. If offered a position and you currently have any affiliation, you must…