Audit a small model through shapes, units and derivatives
Audit a small model through shapes, units and derivatives.
Text lesson
Loading this section…
Use shapes, derivatives, probability and information to reason about models.
Choose any module that helps with your current question. This collection has no required sequence.
26 of 26 modules
Audit a small model through shapes, units and derivatives.
Text lesson
Derive matrix gradients with axes and bias accumulation.
Text lesson
Propagate a scalar derivative at one forward-pass point.
Text lesson
Accumulate every path to a shared parameter.
Text lesson
Use constraints and dual bounds to certify a convex solution.
Text lesson
Check a descent step using gradients and curvature.
Text lesson
Construct binary, count and waiting-time events before choosing a formula.
Text lesson
Build a likelihood estimator and audit its bias, risk and consistency.
Text lesson
Relate entropy, cross-entropy and KL with stable log calculations.
Text lesson
Separate a Jacobian action, a gradient pullback and directional curvature.
Text lesson
Compute a posterior mode and explain its coordinate dependence.
Text lesson
Calculate weighted entropy reduction and explain its distinction from a single observed branch. Diagnose when conditional information differs from marginal dependence.
Text lesson
Handle a kink with subgradients and a proximal step.
Text lesson
Turn density and normal standardization into interval probabilities.
Text lesson
Diagnose a cancellation-prone expression using algebra and rounded intermediates. Interpret finite-difference evidence across step sizes and test points.
Text lesson
Separate missingness assumptions from noisy measurements.
Text lesson
Compute conditionals and moments from a joint model.
Text lesson
Diagnose rank and sensitivity through singular values.
Text lesson
Compare location summaries under a specified contamination experiment and preserve the intended estimand.
Text lesson
Solve a small worst-case objective and state what its guarantee covers.
Text lesson
Distinguish observations, sample means and repeated-sampling uncertainty.
Text lesson
Audit curvature updates, box-bound stationarity and trust-region agreement.
Text lesson
Trace a deliberate stop-gradient boundary.
Text lesson
Translate named tensor axes into an explicit contraction.
Text lesson