AI Research Scientist, Scientific ML

About the role

Western Digital is looking for an AI Research Scientist focused on scientific machine learning, based in Singapore. This is not a generalist AI research role: the candidate studies how physics constraints interact with neural network training and deploys that methodology in real product development, working alongside ML engineers and storage domain experts.

Key responsibilities

• Originate and advance PINNs methodology: design physics-constrained loss function architectures, validate digital twin ML components against domain physics, and deliver validated prototypes with complete technical documentation.
• Lead research in one of three areas: Bayesian deep learning, active learning acquisition functions and ensemble uncertainty methods (Area B); or causal inference and structural causal models for reliability root cause analysis (Area C).
• Design physics-constrained generative models (diffusion models, VAEs) for synthetic data generation and deliver validated training recipes.
• Interface with the storage domain expert to validate physics constraints, publish preprints or patent disclosures, and take part in design reviews as the research methodology authority.

Required profile

• Master's or PhD in AI, ML, Physics, Applied Mathematics or a related field, with a strong scientific computing background.
• Master's: 1-3 years of work or research experience in scientific ML or applied AI; PhD: no minimum experience, peer-reviewed publication or significant open-source scientific ML contribution preferred.
• Demonstrated research output: publication, preprint, thesis chapter or open-source contribution in at least one primary area.
• Ability to turn validated research prototypes into fully documented handoffs.

Required skills

• PyTorch or JAX, at expert, research-level depth.
• PINNs methodology and physics-constrained loss function architectures (required).
• One of: Bayesian deep learning, Bayesian experimental design, active learning acquisition functions, ensemble uncertainty quantification, or structural causal models (one of B or C required).
• Nice to have: diffusion models, graph neural networks, neural ODEs, foundation model fine-tuning, RL for scientific discovery, and a background in materials science, semiconductors or precision products.