Foundation models are changing how science is done, but the way they learn is still poorly understood, and training them on scientific data runs into real limits: scale, cost, and data that can’t leave the site where it was collected. Our group at Argonne National Laboratory works on the methods underneath: the mathematics of how stochastic gradient training behaves, and algorithms that make training efficient, reliable and private across distributed and federated settings. We are looking for a postdoctoral appointee to join this work, from theory through to open-source software that others use.
This Postdoctoral Appointee will be part of the Mathematics and Computer Science Division and will work with partners at universities and other national laboratories and have access to some of the world’s most powerful computing systems. Appointees are encouraged to shape their own research direction and publish at leading machine learning and optimization venues. Postdoctoral appointees are expected to lead papers.
Core responsibilities:
Conducts research on optimization and training dynamics for large-scale machine learning, combining analysis with careful experiments.
Designs, implements and benchmarks new training algorithms for distributed and federated settings, with attention to efficiency, privacy and robustness.
Contributes research software to open-source projects and keeps experiments reproducible.
Publishes and presents results at leading conferences and journals, and contributes to program reports.
Works with collaborators across institutions.
Postdoctoral appointees also help mentor students.
Grows research independence and professional skills such as communication, networking and leadership.
Position Requirements
Required skills:
Ph.D. (completed within the last 0–5 years) in one of the fields above, with a publication record in optimization for machine learning, deep learning theory, or distributed or federated learning
Strong programming skills in Python and PyTorch, including training deep learning models on GPUs
Coursework or research experience in optimization, probability or machine learning
Clear communication and the ability to work in a team spread across several institutions
Ability to model Argonne's core values of impact, safety, respect, teamwork and integrity
Desired skills (nice to have, not required):
Background in stochastic optimization, heavy-tailed noise, or the analysis of loss landscapes and generalization
Experience with federated learning, differential privacy or other privacy-preserving methods
Experience fine-tuning or pretraining foundation models, including parameter-efficient methods like LoRA (low-rank adaptation)
Experience with multi-node distributed training (for example DeepSpeed, PyTorch FSDP or MPI)
Contributions to open-source machine learning software
Job Family
Postdoctoral
Job Profile
Postdoctoral Appointee
Worker Type
Long-Term (Fixed Term)
Time Type
Full time
The expected hiring range for this position is $72,879.00-$121,465.00.
Please note that the pay range information is a general guideline only. The pay offered to a selected candidate will be determined based on factors such as, but not limited to, the scope and responsibilities of the position, the qualifications of the selected candidate, business considerations, internal equity, and external market pay for comparable jobs. Additionally, comprehensive benefits are part of the total rewards package.
Click here to view Argonne employee benefits!
As an equal employment opportunity employer, and in accordance with our core values of impact, safety, respect, integrity and teamwork, Argonne National Laboratory is committed to a safe and welcoming workplace that fosters collaborative scientific discovery and innovation. Argonne encourages everyone to apply for employment. Argonne is committed to nondiscrimination and considers all qualified applicants for employment without regard to any characteristic protected by law.
Argonne…