Find the Minimum
The nouns change — loss, calibration error, portfolio risk — but the procedure is one: pick a number to improve, move through choices, stop when improvement runs out. Once you read it once, you recognise it everywhere.
the skeleton
- 1 objectivePick a single number that should get smaller (or larger).
- 2 moveDecide which direction in the choice space lowers the objective.
- 3 stepDecide how far to move in that direction.
- 4 stopPick a rule for when further moves stop helping.
instances · 3
ml / dl · gradient-descent
Gradient Descent
objective training loss L(w)
stops when gradient norm small or held-out loss flat.
ml / dl · model-calibration
Model Calibration
objective log-loss on validation set
stops when validation log-loss stops improving as temperature T changes.
finance · portfolio-risk
Portfolio Risk
objective portfolio variance σ²
stops when KKT conditions met — no feasible move lowers risk further.
leans on
walk the instances
Finding the Minimum →
One skeleton — objective, move, step, stop — under three applications across ML and finance.