Principal Applied Scientist - Foundation Models, Agents & Decision Intelligence

About the Role

Microsoft Advertising is pioneering the development of advanced AI systems aimed at understanding advertiser behavior, detecting anomalies, and identifying emerging threats. We seek a Principal Applied Scientist with a robust mathematical, statistical, and core machine learning background to drive innovations in foundation models related to behavior, content, entities, and risk comprehension. The role focuses on anomaly detection and threat modeling for evolving abuse patterns, uncertainty modeling in decision-making processes and agents that support case investigations and automated or human decision workflows. Responsibilities include rigorous evaluation of models and decision systems, handling large-scale diverse data, and ensuring solutions are adaptable across products, markets, and adversarial contexts.

Responsibilities

• Lead scientific projects in areas such as foundation models, behavioral modeling, anomaly detection, threat modeling, and agentic systems development.
• Engineer scalable learning systems that analyze entities, content, relationships, and behaviors over time while identifying multiple types of risks.
• Develop techniques to assess and propagate uncertainty across models, cascades, agents, evidence collection, automated decisions, and human reviews.
• Leverage uncertainty, confidence, and risk factors to decide automation thresholds, evidence gathering, invoking advanced systems, abstention, or escalation for expert review.
• Convert insights from adversarial and threat models into actionable data strategies, model architectures, and comprehensive evaluation frameworks.
• Advance agent training and post-training processes for systems that investigate complex cases by using tools and evidence to achieve well-founded outcomes.
• Address machine learning challenges such as distribution shifts, sparse or delayed labels, noisy supervision, class imbalance, selective observation, and adaptive adversaries.
• Translate research findings into dependable, efficient production-grade AI capabilities within Microsoft Advertising.
• Provide leadership, mentor team members, and shape long-term AI architectures focused on trust and safety.

Qualifications

• Academic background at the Bachelor's, Master's, or Doctorate level in Computer Science, Mathematics, Statistics, Electrical Engineering, Operations Research or a related quantitative discipline along with relevant research or industry experience.
• Strong expertise in probability, statistics, linear algebra, optimization, numerical methods, experimental design, and statistical decision theory.
• In-depth knowledge of modern machine learning fields such as foundation or representation learning, behavioral and temporal modeling, and anomaly detection.
• Demonstrated experience with post-training and evaluation of large-scale models involving multiple billion parameters.
• Competence in modeling uncertainty within production-level decision systems.
• Proven capability to analyze and model threat and abuse scenarios.
• Strong proficiency in Python coding and experience with ML frameworks like PyTorch, JAX, TensorFlow or comparable technologies.
• Track record of driving scientific projects from concept through experimentation, production deployment, and measurable impact.
• Proven technical leadership including scientific direction, architectural decisions, mentorship, and cross-disciplinary collaboration.

Preferred Qualifications

• Experience with tool-using agents, retrieval techniques, agent post-training, reward modeling, or trajectory evaluation methods.
• Background in trust and safety domains such as fraud detection, abuse prevention, cybersecurity, content moderation, account integrity, or policy enforcement.
• Familiarity working with temporal, multimodal, heterogeneous, or graph-structured datasets.
• Strong publication record or practical achievements in machine learning topics including agents, anomaly detection, probabilistic modeling, adversarial machine learning, multimodal learning, or trust and safety.

Additional Information

This role involves full ownership of scientific initiatives, requiring both theoretical expertise and practical implementation skills. Microsoft encourages equal employment opportunities regardless of personal characteristics and offers accommodations for applicants needing support during the application or hiring process.