Job DescriptionWe have an opening for a
Postdoctoral Researcher in Explainable AI to contribute to fundamental R&D on understanding what modern AI models learn and how that knowledge is represented internally. As foundation models and deep surrogates inform consequential scientific and national security decisions, domain experts need to inspect, validate, and steer model internals, making interpretability as much a human-AI collaboration problem as a modeling one. Your work will focus on recovering human-meaningful structure from learned representations, including sparse decompositions of activations, concept discovery, mechanistic analysis, and causal intervention, and on the interactive interfaces and evaluation methodology that let experts interrogate that structure and the given explanations. Applications area includes but not limited to multimodal SciML models and deep surrogates for various simulations. This position will be in the Machine Intelligence Group in the Center for Applied Scientific Computing (CASC) Division within the LLNL Computing Directorate.
This position offers a hybrid schedule, blending in-person and virtual presence. You will have the flexibility to work from home one or more days per week.
Essential Duties- Develop and evaluate methods for interpreting the internal representations of deep models, including sparse decompositions of activations, concept extraction, and representation steering.
- Design human-in-the-loop workflows that let domain experts explore, validate, and correct discovered concepts, and evaluate those workflows with real users.
- Establish rigorous evaluation methodology for interpretability claims, i.e., faithfulness, stability, and causal grounding.
- Research, design, implement, and apply advanced machine learning methods for multiple applications in a collaborative scientific environment.
- Conduct cutting-edge machine learning research effectively and independently.
- Actively participate with project scientists and engineers in defining, planning, and formulating experimental, modeling, and simulation efforts for complex problems stemming from national security applications.
- Propose and implement advanced analysis methodologies, collect and analyze data, and document results in technical reports and peer-reviewed publications.
- Contribute to grant proposals and collaborate with others in a multidisciplinary team environment, including academic and industrial partners, to accomplish research goals.
- Pursue independent (but complementary) research interests and interact with a broad spectrum of scientists internal and external to the Laboratory.
- Perform other duties as assigned.
Qualifications- Recent Ph.D. in Computer Science, Machine Learning, Applied Mathematics, Statistics, Human-Computer Interaction, or a related field.
- In-depth knowledge in explainable AI and related topics, demonstrate relevant experiences and corresponding publications.
- Demonstrated research experience in explainable or interpretable AI, representation learning, mechanistic interpretability, concept-based explanation, or visual analytics for machine learning.
- Experience developing and applying deep learning methods at medium to large scale using modern libraries such as PyTorch or JAX.
- Demonstrated research productivity, as documented by publications, reports, presentations, and/or open-source software in relevant venues (NeurIPS, ICML, ICLR, CVPR, ACL, IEEE VIS, CHI, JMLR, etc.).
- Experience with scientific programming in the Python ecosystem, and demonstrated ability to obtain substantial domain knowledge in fields of application in order to communicate effectively with subject matter experts.
Desired Qualifications- Experience with sparse autoencoders, transcoders, or related feature-learning methods applied to the activations of large pretrained models.
- Experience analyzing or intervening on the internal representations of trained models, such as probing for encoded properties, steering or editing activations to alter behavior, or attributing outputs to internal components.
- Experience connecting interpretability to uncertainty quantification, robustness, calibration, or AI safety and assurance evaluation.
- Experience with high-performance computing, GPU programming, parallel programming, cloud computing, and/or related methods including running numerical simulations of complex workflows.
- Demonstrated technical leadership in fields related to machine learning, such as mentorship or managing teams.
- Experience or interest in scientific applications, such as, material science, climate science, etc.
Pay Range$143,328 Annually
Additional InformationAll your information will be kept confidential according to EEO guidelines.
Position InformationThis is a Postdoctoral appointment with the possibility of extension to a maximum of three years, open to those who have been awarded a PhD at time of hire date.
Security ClearanceNone required. However, if your assignment is longer than 179 days cumulatively within a calendar year, you must go through the Personal Identity Verification process. This process includes completing an online background investigation form and receiving approval of the background check.
