Join Microsoft Advertising's cutting-edge AI team, pioneering the next generation of systems designed to understand advertiser behavior, detect anomalies, and combat emerging threats. This role focuses on advancing foundational models for understanding behavior, content, entities, and risk. You will also work on anomaly detection, threat modeling, decision uncertainty, and developing intelligent agents for case investigation and decision support.

This hands-on scientific position involves end-to-end ownership. You'll be responsible for problem formulation, model development, large-scale training, evaluation, productionization, and ultimately, driving measurable product impact.

Define and lead scientific initiatives in areas such as foundation models, behavioral modeling, anomaly detection, and agentic systems. Develop scalable learning systems capable of understanding entities, content, relationships, and temporal behavior while identifying risks. Create methods to model and propagate uncertainty across various components, influencing automation and escalation decisions.

Translate threat models into actionable data strategies and model architectures. Enhance the training and evaluation of tool-using agents for complex case investigations. Address challenging learning scenarios like distribution shift and noisy labels. Drive the transition of scientific advancements into reliable production capabilities and provide technical leadership and mentorship.

A Bachelor's, Master's, or Doctorate degree in a quantitative field such as Computer Science, Mathematics, or Statistics is required, along with relevant industry or research experience. A strong foundation in probability, statistics, linear algebra, optimization, and experimental design is essential.

Deep expertise in modern machine learning, including foundation models, behavioral modeling, and anomaly detection, is crucial. Proven experience in post-training and evaluating large-scale models, alongside modeling uncertainty in production systems, is expected. Proficiency in Python and ML frameworks like PyTorch, JAX, or TensorFlow is necessary, as is a track record of taking scientific ideas from concept to production.