← Question bank
90 questions · full solutions
Machine Learning
The quant-researcher round: bias and variance, leakage-proof validation, regularization, and the finance-specific failure modes most ML courses never mention.
- 01Bias, Variance, and the Noise FloorMedium · Private
- 02Why Your Backtest LiesMedium · Private
- 03Ridge vs Lasso, and Why Quants Love RidgeMedium · Private
- 04Log-Loss from First PrinciplesMedium · Private
- 05Two Roads to the Same LineMedium · Private
- 06Neighbors Who Live Far AwayHard · Private
- 07Margins and HingesMedium · Private
- 08The Kernel Trick in One ParagraphMedium · Private
- 09Naive Bayes Counts TwiceEasy
- 10Why One Tree Is a GambleEasy
- 11The Forest's Variance FloorMedium · Private
- 12Boosting Is Gradient Descent in Function SpaceHard · Private
- 13What the λ and γ Are DoingHard · Private
- 14Importance Is Not ImportanceMedium · Private
- 15AUC by EnumerationMedium · Private
- 16The 99% Accurate Crash DetectorMedium · Private
- 17Right Order, Wrong ProbabilitiesHard · Private
- 18The Cost of Saying NeverMedium · Private
- 19Accuracy TheaterEasy
- 20When Gradient Descent ExplodesMedium · Private
- 21The Useful Noise in SGDMedium · Private
- 22Which Objectives Are Convex?Easy
- 23One Step of NewtonMedium · Private
- 24Why Deep Sigmoids Stop LearningEasy
- 25Universal Approximation, Limited ComfortHard · Private
- 26Dropout: An Ensemble in DisguiseMedium · Private
- 27Batch Norm, and the Leak Nobody MentionsMedium · Private
- 28Past the Interpolation ThresholdHard · Private
- 29Two Eigenvalues, One DirectionMedium · Private
- 30What k-Means Actually OptimizesEasy
- 31PCA Meets the Cross-SectionMedium · Private
- 32The Label Is the StrategyHard · Private
- 33A Thousand Rows, Two Hundred FactsHard · Private
- 34Trained on a World That EndedHard · Private
- 35The Best of a Hundred NothingsHard · Private
- 36Gross Genius, Net RealityMedium · Private
- 37One Model for Direction, Another for ConvictionHard · Private
- 38Models That AgeMedium · Private
- 39Count the WeightsEasy
- 40One Number from Four CellsEasy
- 41The Dummy Variable TrapEasy
- 42Ridge, One Eigenvalue at a TimeMedium · Private
- 43The Prior That PullsMedium · Private
- 44When the Noise Isn't FlatMedium · Private
- 45One Split, By HandMedium · Private
- 46The Word That Never AppearedMedium · Private
- 47Lasso's Knife, Ridge's SqueezeMedium · Private
- 48Ten Flavors of MomentumMedium · Private
- 49Fifty Times the VarianceMedium · Private
- 50Only the Tails Get TradedMedium · Private
- 51Confidence, AuditedMedium · Private
- 52Anatomy of One HeadMedium · Private
- 53The Unit That Stopped AnsweringMedium · Private
- 54Bias and Variance, With Actual NumbersHard · Private
- 55Deliberately Wrong on PurposeHard · Private
- 56Two Copies, One CoefficientHard · Private
- 57Every Penalty Is a PriorHard · Private
- 58The Best Possible ShrinkageHard · Private
- 59The Likelihood That Never Stops ClimbingHard · Private
- 60Trained on a Balanced LieHard · Private
- 61Manufacturing Skill from NothingHard · Private
- 62Purge, Then EmbargoHard · Private
- 63The Score You Tuned Into ExistenceHard · Private
- 64Can Your Validation Set Even Tell?Hard · Private
- 65Twenty Tickets in the Sharpe LotteryHard · Private
- 66Your Labels Remember Each OtherHard · Private
- 67How Much Data Do You Actually Have?Hard · Private
- 68Two Regimes, Zero SignalHard · Private
- 69Left Out of the BagEasy
- 70The Floor No Model BeatsMedium · Private
- 71The Gradient Is the MissMedium · Private
- 72Coverage Without Believing the ModelMedium · Private
- 73Prediction as Weighted SimilarityMedium · Private
- 74Why 0.5 Is Almost Never the ThresholdMedium · Private
- 75Two Rounds of AdaBoost, By HandHard · Private
- 76One Honest Step of EMHard · Private
- 77One Attention Lookup, By HandHard · Private
- 78The Chain Rule, AuditedHard · Private
- 79Ten Layers Later, the Signal Is GoneHard · Private
- 80Betting on Experts with Multiplicative WeightsHard · Private
- 81Taking the Brier Score ApartHard · Private
- 82The Importance That Halved ItselfHard · Private
- 83The R-Squared You Get for FreeHard · Private
- 84Epochs Are a Regularization DialHard · Private
- 85The Feature That Contains Its Own AnswerHard · Private
- 86Training Error Owes You MoneyHard · Private
- 87What a Tiny R-Squared Is WorthHard · Private
- 88The Loss That Forecasts a QuantileHard · Private
- 89Find the Three LeaksHard · Private
- 90The Category Trick That CheatsHard · Private