Skip to content

tutorial_outline.md

Latest commit

 

History

History
78 lines (50 loc) · 2.52 KB

tutorial_outline.md

File metadata and controls

78 lines (50 loc) · 2.52 KB
Raw

Tutorial Outline

This outline summarizes the completed ASD agent tutorial and the BO extension.

Part 1: Virtual Laboratory

  • Define GA and NGA surfaces.
  • Explain saturating precursor and coreactant response.
  • Show nucleation delays, inhibitor blocking, temperature response, and noise.
  • Emphasize that the model is educational and non-predictive.

Notebook: notebooks/01_understanding_the_virtual_lab.ipynb

Part 2: Baseline Optimization

  • Run deterministic rule-based search.
  • Compare random and grid baselines.
  • Inspect ledgers, success criteria, and failure categories.

Notebook: notebooks/02_rule_based_optimization.ipynb

Part 3: LLM Agent Loop

  • Explain strict tools: propose_experiments and finish_optimization.
  • Separate Codex as software-building agent from the scientific optimization agent.
  • Demonstrate no-API deterministic alternatives.

Notebook: notebooks/03_llm_agent.ipynb

Part 4: Legacy Benchmarking

  • Compare random, grid, rule-based, and optional LLM methods.
  • Generate summary plots and discuss failure modes.

Notebook: notebooks/04_benchmark_and_failure_analysis.ipynb

Part 5: Stage 1 Saturation Learning

  • Define one-dimensional saturation-process families.
  • Compare fixed grid, generic GP, and physics-informed GP.
  • Discuss t95 estimation, cumulative dose, calibration, and misspecification.

Notebook: notebooks/05_bayesian_active_learning_for_saturation.ipynb

Part 6: Stage 2 Constrained MOBO

  • Define objectives and constraints for two-surface ASD optimization.
  • Explain why selectivity alone is insufficient.
  • Compare constrained MOBO against random, grid, and rule-based search.

Notebook: notebooks/06_constrained_multiobjective_asd_optimization.ipynb

Part 7: Hybrid LLM-BO Agent

  • Use BO as the numerical optimizer.
  • Use the LLM to inspect history, request candidates, explain choices, and propose bounded interventions.
  • Run the fake LLM by default.

Notebook: notebooks/07_hybrid_llm_bo_agent.ipynb

Part 8: Research Harness

  • Configure paired smoke, pilot, and paper profiles.
  • Generate paired statistical summaries and failure taxonomies.
  • Keep hypotheses separate from conclusions.

Notebook: notebooks/08_research_benchmark_and_statistics.ipynb

Part 9: Manual Laboratory Handoff

  • Export validated manual lab plans.
  • Import completed measurements.
  • Continue BO from accepted human-operated measurements.
  • Document remaining work before any real laboratory validation.

Primary docs: docs/lab_validation_protocol.md and model cards under docs/model_cards/.