Part I
Learning
How models learn representations, acquire knowledge, and build useful models of the world.
Topics
- Transformers and emerging architectures
- Scaling laws and pre-training
- Multimodality
- Long-context, retrieval, and memory
- World models
MLx FRONTIER AIOXFORD MACHINE LEARNING SCHOOL — OXML 2027
A four-part curriculum on models that learn, reason, act, and evolve.
Frontier AI is moving beyond prediction into reasoning and action. OxML 2027 is built around that shift.
Across four connected parts, we cover how models learn representations and world knowledge; how they reason and plan with more compute at inference time; how they act through tools, agents, and environments; and how systems can improve, evaluate, and evolve over time.
This is rigorous training for postgraduate students, researchers, and engineers who want depth — not a survey tour.
Lectures are taught by researchers publishing at NeurIPS, ICML, ICLR and building systems at the frontier of capability.
How models learn representations, acquire knowledge, and build useful models of the world.
Topics
How models plan and solve complex problems using more compute and better inference-time strategies.
Topics
How models use tools, take action, and interact with environments to achieve goals.
Topics
How frontier systems are evaluated and aligned — and how self-improvement, continual learning, and open-endedness shape where capabilities are heading.
Topics
OxML 2027 — Oxford
See how the programme runs today with OxML 2026 — or check back when 2027 applications open.