Expand description
ML-specific nullification hypotheses.
Functions§
- accuracy_
baseline - H₀-MULTI-01: Accuracy >= oracle baseline.
- gradient_
bounded - H₀-TRAIN-03: Gradient norms remain bounded.
- latency_
sla - H₀-PRED-03: Latency within specified SLA.
- loss_
stability - H₀-TRAIN-02: Loss monotonically decreases (no spikes).
- params_
finite - H₀-TRAIN-05: Model parameters remain finite.
- prediction_
determinism - H₀-PRED-01: Predictions are deterministic for temperature=0.
- statistical_
significance - H₀-MULTI-05: Results are statistically significant (5 runs, t-test).
- training_
determinism - H₀-TRAIN-01: Training converges identically across runs with same seed.