AI Research Scientist, Scientific ML

WD is building the infrastructure behind the AI-driven data economy. As AI scales, so does data. Every interaction, every model, and every system generates data that must be stored, managed, and made accessible over time. That’s where we come in. We combine deep engineering expertise with global-scale manufacturing to deliver the storage systems that make AI possible, powering hyperscale data centers, cloud platforms, and enterprise infrastructure worldwide. This is real systems work, tackling some of the hardest challenges in technology today. We are looking for people who want to build, solve, and operate at that level. Join us and let’s shape the future of data.

About This Role — The Mission

This is not a generalist AI research role. We are seeking a researcher who has deeply considered how physics constraints interact with neural network training and wishes to apply that methodology to real product development challenges, rather than only validating on benchmark datasets. You will be the architect of what the ML engineers build. Your acquisition functions will drive real laboratory experiments, and your PINNs methodology will run within product development. The work originating here will be tested against physical ground truth in ways most academic scientific ML researchers never access.

Key Areas of Focus

• Area A — Scientific ML & PINNs Methodology Origination: Originating and advancing PINNs methodology by designing physics-constrained loss function architectures, validating digital twin ML components against domain physics (with storage domain experts), and delivering validated prototypes with complete technical documentation. As the sole PINNs methodology originator on the team, this capability cannot be delegated or substituted.
• Area B — Uncertainty Quantification & Bayesian Experimental Design: Researching Bayesian deep learning, designing active learning acquisition functions, ensemble uncertainty methods, and Bayesian experimental design frameworks for autonomous experiment selection. Transferring validated acquisition function designs for active learning pipeline integration.
• Area C — Causal ML & Reliability Modeling: Owning the causal inference framework product development for reliability root cause analysis, including structural causal model (SCM) design, causal discovery, and causal intervention planning for product development improvement.

Synthetic Data Methodology

Designing physics-constrained generative model approaches (such as diffusion models and VAEs) for synthetic data generation. This includes delivering a validated methodology and training recipes for operational pipeline implementation.

Intellectual Property & Domain Interface

Demonstrating strong research output through preprints or patent disclosures. Interfacing with storage domain experts to validate physics constraints before deployment. Producing validated research prototypes with complete technical documentation for team handoff, and participating in design reviews as the research methodology authority.

Qualifications

Education

• Master's or PhD in Artificial Intelligence, Machine Learning, Physics, Applied Mathematics, or a related field. A strong focus on AI/ML research and a background in scientific computing are required.

Experience and Required Skills

For Master's degree holders: 1–3 years of work or research experience in scientific ML or applied AI roles.

For PhD holders: Open—no minimum work experience is required; research depth is the primary criterion. Must demonstrate peer-reviewed publications (NeurIPS, ICML, ICLR, AAAI, Nature MI, or domain-specific venues), strong PhD research, or significant open-source scientific ML contributions.

• Technical Mastery: Expert skill level in PyTorch or JAX, demonstrated through deep research-level implementation capability.
• Core Requirement (Area A): Proficiency in PINNs methodology design, physics-constrained loss function architecture, and digital twin modeling. This is the most critical capability sought on the team.
• Specialization Requirement (Area B or C): Expertise in either:
  • Bayesian deep learning, Bayesian experimental design, active learning acquisition function design, or ensemble uncertainty quantification (Area B).
  • Structural causal models (SCM), causal discovery, or causal inference for reliability/yield root cause analysis (Area C).
• Research Deliverables: Demonstrated research output via publication, preprint, PhD thesis chapter, or significant open-source scientific ML contribution in at least one primary area.
• Technical Handoff: Ability to produce validated research prototypes complete with thorough technical documentation suitable for engineering handover.

Good to Have Skills

• Diffusion models (DDPM, conditional diffusion) for physics-constrained synthetic data generation.
• Graph neural networks (GNN) for materials property prediction or failure propagation modeling.
• Neural ODEs for dynamic systems and degradation trajectory modeling.
• Fine-tuning foundation models for domain adaptation in scientific tasks.
• Reinforcement Learning (RL) for scientific discovery and exploration strategies in experimental search spaces.
• Prior top-venue publication (NeurIPS / ICML / ICLR / Nature MI) is a strong bonus signal.
• Background in materials science, semiconductor, or precision product development.