TUTOR

M25 · Level V — Frontier Intelligence & Research

Scientific Machine Learning

Build scientific machine learning systems that preserve, learn from, or explicitly interrogate mathematical structure in physical and scientific problems. The learner must move from a physical law or measurement process to a validated computational model, not merely fit a predictor to a table.

Topics

  • Physics
  • Data
  • Dynamics
  • Inference
  • Differentiation
  • Operators
  • Uncertainty
  • Discovery

Technical concepts

  • 01Learning Contract: Question -> Model -> Evidence
  • 02Scientific Measurement and Data-Generating Processes
  • 03Dimensional Analysis, Scaling and Non-dimensionalisation
  • 04Dynamical Systems: State, Flow and Observation
  • 05ODE Initial-Value Problems and Numerical Integration
  • 06Stability, Convergence and Error Separation
  • 07PDE Thinking: Fields, Boundaries and Conservation
  • 08Discretisation: From Continuous Laws to Computable Objects
  • 09Numerical Stability as a Learning Constraint
  • 10Inverse Problems: Infer Causes from Observations
  • 11Ill-Posedness, Conditioning and Identifiability
  • 12Regularisation: Priors as Computation and Science
  • 13Parameter Estimation and System Identification
  • 14Bayesian Scientific Inference
  • 15Uncertainty Propagation Through Scientific Pipelines
  • 16Ensembles, Calibration and Scientific Decision Thresholds
  • 17Surrogate Modelling and Expensive Simulators
  • 18Gaussian Processes as a Scientific Baseline
  • 19Operator Learning: Learn Maps Between Function Spaces
  • 20Physics-Informed Learning: Residuals, Constraints and Data
  • 21Why Physics-Informed Learning Can Fail
  • 22Differentiable Simulation and End-to-End Sensitivity
  • 23Adjoint Reasoning and Inverse Design
  • 24Conservation Laws and Structure-Preserving Learning
  • 25Symmetry, Invariance and Equivariance
  • 26Scientific Graphs, Meshes and Geometric Data
  • 27Data Assimilation: Fuse Model and Observation
  • 28Sequential Bayesian Estimation and Sensor Fusion
  • 29Hybrid Scientific Models
  • 30Multi-Fidelity Modelling
  • 31Active Learning for Scientific Experiments
  • 32Scientific Experiment Design and Identifiability
  • 33Causal Structure in Scientific Modelling
  • 34Scientific Discovery Loops
  • 35Time-Series Learning and System Identification
  • 36Reduced-Order Modelling and Latent Dynamics
  • 37Digital Twins as Scientific Systems
  • 38Verification, Validation and Scientific Credibility
  • 39Validation of Uncertainty and Extrapolation
  • 40Model Discrepancy and Error-Aware Hybridisation
  • 41Scientific Computing Stack and Reproducible Runtimes
  • 42Reproducibility, Provenance and Research Hygiene
  • 43Technical Session - Scientific ML in Real Systems
  • 44Frontier Bridges: Foundation Models, World Models and Quantum Computing
  • 45Core Methods and Tool Checklist

Practical implementation

  • 01Lab 1 - Build a Scientific Data Pipeline
  • 02Lab 2 - Validate an ODE Solver
  • 03Lab 3 - Solve a Parameter Inverse Problem
  • 04Lab 4 - Build a Surrogate for an Expensive Simulator
  • 05Lab 5 - Physics-Informed Boundary-Value Learning
  • 06Lab 6 - Differentiate Through a Simulator
  • 07Lab 7 - Data Assimilation and Sensor Fusion
  • 08Lab 8 - Learn a Hybrid Model
  • 09Lab 9 - Scientific Discovery Mini-Loop

Tools and technologies

  • Linux VM/container
  • numerical simulation
  • differentiable programming
  • inverse problems surrogate models
  • operator learning
  • uncertainty quantification
  • scientific validation
  • reproducible research

Project

Lab 9 - Scientific Discovery Mini-Loop

Expected learning outcomes

  • 01At completion, the learner should be able to enter an unfamiliar scientific domain and build a disciplined computational loop without mistaking a fashionable architecture for a scientific method.
  • 02A scientific model is acceptable only when its assumptions, equations, data-generating process, numerical approximation, uncertainty, and validation evidence are explicit enough to reproduce and challenge the result.

Assessment

  • 01Concept / mathematics exams — 20%
  • 02Laboratory work — 25%
  • 03Daily practice — 15%
  • 04Research reproduction — 10%
  • 05Viva — 10%
  • 06Capstone — 20%

Source document: TUTOR Frontier Module 25 Scientific Machine Learning