M29 · Level V — Frontier Intelligence & Research
AI Research Engineering
Train the learner to convert an open intelligent-systems question into a defensible research program: precise problem statement, mechanism hypothesis, controlled experiment, reproducible implementation, evidence, interpretation and research communication.
Topics
Technical concepts
- 01Research as an Evidence Pipeline
- 02Problem Selection and Scope
- 03Literature Search as Technical Investigation
- 04Claim, Contribution and Novelty
- 05Hypotheses and Mechanistic Predictions
- 06Variables, Controls and Experimental Units
- 07Baselines and Counterfactual Baselines
- 08Dataset Construction and Data Provenance
- 09Benchmark Design and Metric Semantics
- 10Measurement Error and Instrumentation
- 11Statistical Power and Sample Size
- 12Randomness, Seeds and Replication
- 13Confidence Intervals and Uncertainty Reporting
- 14Multiple Comparisons and Researcher Degrees of Freedom
- 15Experimental Logging and Configuration
- 16Reproducible Environments
- 17Data and Artifact Versioning
- 18Reproduction of Published Results
- 19Replication vs Reproduction
- 20Ablation Studies
- 21Intervention and Counterfactual Testing
- 22Scaling Laws and Resource-Performance Curves
- 23Compute Budgets and Research Efficiency
- 24Hyperparameter Search as an Experimental Variable
- 25Benchmark Contamination and Leakage
- 26Robustness and Distribution Shift
- 27Out-of-Distribution and Uncertainty Experiments
- 28Error Taxonomies and Failure Clustering
- 29Interpretability as a Research Instrument
- 30Probing Learned Representations
- 31Activation and Parameter Interventions
- 32Representation Geometry and Symmetry
- 33Optimisation Dynamics as a Research Object
- 34Training Dynamics and Phase Changes
- 35Curriculum, Data Ordering and Learning Pathways
- 36Architecture Search and Structural Hypotheses
- 37Generative Model Research
- 38Language Model Research
- 39Vision and Multimodal Research
- 40Agent and Tool-Use Research
- 41World Models and Model-Based Research
- 42Scientific ML and Differentiable Simulation
- 43Reinforcement Learning Research
- 44Robotics and Autonomous Systems Research
- 45Performance Engineering as Research
- 46Research Software Architecture
- 47Continuous Integration for Research
- 48Experiment Graphs and Lineage
- 49Peer Review Readiness
- 50Research Decision and Next-Experiment Design
- 51Technical Session - How Frontier Research Becomes an Engineering Artifact
- 52Cross-Domain Bridge - Quantum Methods in Research Engineering
- 53Research Methods and Tool Checklist
Practical implementation
- 01Lab 1 - Research Question to Hypothesis
- 02Lab 2 - Literature Map and Baseline Matrix
- 03Lab 3 - Reproduce a Published Result
- 04Lab 4 - Ablation and Mechanistic Study
- 05Lab 5 - Scaling and Compute-Efficiency Study
- 06Lab 6 - Representation or Behaviour Intervention
- 07Lab 7 - Research CI and Artifact Pipeline
- 08Lab 8 - Independent Replication and Stress Campaign
- 09Lab 9 - Research Note and Next-Experiment Defense
Tools and technologies
Project
Lab 9 - Research Note and Next-Experiment Defense
Expected learning outcomes
- 01A research claim is only as strong as the chain connecting the question, assumptions, method, measurements, controls and evidence. A working implementation is not by itself a scientific result.
Assessment
- 01Research theory and derivations — 20%
- 02Laboratories — 25%
- 03Reproduction / replication — 15%
- 04Mechanistic study — 15%
- 05Research note — 15%
- 06Viva — 10%
Source document: TUTOR Frontier Module 29 AI Research Engineering