Postdoctoral Research Associate - AI Science

Overview

The Analytics and AI at Scale (AAIMS) group under Advanced Technology Section (ATS) of NCCS is hiring two postdoctoral research associates to push the frontier of agentic AI for science, scientific reasoning, federated & collaborative learning, and reinforcement learning (RL) for self‑improving models in the context of leadership scientific workflows and applications.

• Agentic AI for Science:
  Autonomous and tool‑using agents for experiment design, simulation steering, data collection, and lab/compute orchestration; planning and memory; multi‑agent collaboration.
• Scientific Reasoning:
  Program/path‑of‑thought, tool‑augmented and retrieval‑augmented reasoning; uncertainty quantification and calibrated decisions.
• RL & Self‑Improving Models: RLHF/RLAIF, online RL, self‑play, open‑ended discovery, reward modeling, curriculum/active learning, data selection, iterative post‑training, safety alignment and guardrails.
• Foundation Models for Science at Scale:
  Pretraining, instruction tuning, continued pretraining, Mixture‑of‑Experts; distributed training/inference (e.g. FSDP, Deep Speed, Megatron‑LM, tensor/sequence parallelism); scalable evaluation pipelines for reasoning and agents.
• Federated & Collaborative Learning:
  Cross‑silo training across institutions and facilities; privacy‑preserving learning (secure aggregation, differential privacy, MPC/HE); personalization under heterogeneity; governance‑aware data/model sharing; collaborative evaluation.

• Conduct and publish original research on AI for science at scale on leadership‑class systems.
• Design, implement, and benchmark large‑scale training and post‑training pipelines (including distributed data/compute and evaluation harnesses).
• Collaborate with domain scientists and external partners; co‑develop end‑to‑end AI workflows that demonstrably accelerate scientific discovery.
• Architect and operate federated and collaborative learning experiments spanning multiple sites (national labs, universities, industry) with secure aggregation, heterogeneity‑aware scheduling, robust and efficient training and fine tuning.
• Contribute to open‑source software, datasets, and standardized evaluation suites; mentor interns and students.
• Communicate results through papers, artifacts, and presentations at top‑tier venues.

• Ph.D. in Computer Science, Computer Engineering, a physical/computational science discipline (e.g., physics, chemistry, materials science, climate, biology), or a closely related field (earned within the last 5 years or expected before the start date).
• Demonstrated research in at least one: agentic AI/LLM agents, scientific reasoning or tool‑augmented LLMs, RL (RLHF/RLAIF/online RL), or foundation models for science.
• Software engineering skills (Python) and experience with modern DL stacks (PyTorch) and multi‑GPU training.
• Evidence of ability to conduct independent research and publish in peer‑reviewed venues.

• Publications in leading venues (e.g., NeurIPS, ICML, ICLR, AAAI, CVPR/ICCV/ECCV, ACL/EMNLP, MLSys/SC/HPDC).
• Hands‑on with distributed training/inference (FSDP, Deep Speed, Megatron‑LM), accelerator programming, and large‑scale data pipelines.
• Experience building agents that use tools/APIs (e.g., code interpreters, simulation frameworks, databases, lab instruments) and evaluation for long‑horizon tasks.
• Experience with RL and post‑training (reward modeling, preference learning, offline/online RL, self‑play, curriculum learning, data selection).
• Experience with federated learning (cross‑silo FL, Fed Avg/Fed Prox/personalization), privacy‑preserving methods (secure aggregation, differential privacy) and collaborative training/evaluation across heterogeneous data and compute.
• Background in one or more scientific domains (materials, chemistry, climate, fusion, biology) and/or scientific software ecosystems.
• Strong communication skills and a collaborative mindset in cross‑disciplinary teams.

Applicants cannot have received their Ph.D. more than five years prior to the date of application and must complete all degree requirements before starting their appointment. The appointment length will be up to 24 months with the potential for extension. Initial appointments and extensions are subject to performance and availability of funding.

Please submit three letters of reference when applying to this position. You may upload these directly to your application or have them sent to Postdoc recruitment with the position title and number referenced in the subject line.

ORNL is an equal opportunity employer. All qualified applicants, including individuals with disabilities and protected veterans, are encouraged to apply. UT‑Battelle is an E‑Verify employer.