Protocol
Diagnose Neural Training Failures
Write a training incident report with competing explanations. Demonstrate which measurement falsifies each explanation before choosing a fix.
35–60 active hours60–90 min sessionsIntermediate
Mastery contract
Core protocol · Standard 1.0 · Write a training incident report with competing explanations. Demonstrate which measurement falsifies each explanation before choosing a fix.
Loading your learning records…
Reasoning — not yet passed
Explain initialization scaling and predict the signature of vanishing gradients and overfitting before observing runs.
Reliable implementation — not yet passed
Instrumentation records the relevant state without retaining an unbounded autograd graph.
Reproducible experiment — not yet passed
All 12 intervention runs use matched data/compute and include unsuccessful interventions in the report.
Defense and handoff — not yet passed
Diagnose an unseen broken configuration, propose one intervention and justify it from measured evidence.
All four criteria must pass. Activity completion does not satisfy them.
An independent reviewer reruns your work and varies something you did not rehearse: a fresh input, a different working directory, or a declared edge case.
Pilot decisions are provisional, not external credentials. For an appeal or sensitive artifact, contact your designated pilot operator with the submission ID. Export your assessment history.
What you will do
6 total
Derive how variance propagates through a linear layer and explain initialization scaling; distinguish decoupled weight decay from an L2 penalty under adaptive updates.
Instrument an MLP with loss, gradient norm, activation distribution and learning-rate logs; add normalization and regularization switches.
Run 4 preregistered interventions one at a time with 3 seeds each; compare curves, held-out loss and work spent rather than selecting the best run.
Debug a vanishing-gradient configuration and a dropout-enabled validation loop; isolate them with tiny-set and eval-mode checks.
Write a 500–900 word technical report linking the reasoning, code, measurements, and limitations; include the project decision below.
Assess all 4 mastery criteria against saved artifacts, request an independent review, and repeat each failed criterion on a fresh example.
Essentials
A reading session alone never satisfies a mastery criterion.
Store source references, assumptions, and artifact paths beside every result.
Keep an untouched check case that differs from the worked example.
Record all attempts, including failures and results that contradict your prediction.
Use the stated comparison conditions; document every deviation before drawing a conclusion.
Protocol guardrails
Do
- +State the expected result before running the comparison.
- +Keep one minimal reproducible failing case when debugging.
- +Record environment versions and the exact command used.
- +Compare explanations with saved intermediate values.
- +Ask a reviewer to challenge the weakest assumption.
Don't
- ×Do not copy a worked solution and present it as an independent implementation.
- ×Do not tune against held-out evaluation outcomes.
- ×Do not report only the best seed or discard inconvenient runs.
- ×Do not equate elapsed hours or a completed run with a passed assessment.
- ×Do not conceal reduced-scale experiments behind claims about the original full-scale result.
Protocol authorship
Written by NuthinButta as instructional design. The teaching sources are the official references linked in each milestone; the exercises, workload and pass thresholds are ours, not their authors'.