Quant interview questions

1,322 questions · page 25 of 34

• A softmax output layer receives logits (2,1,0).Machine learning · Foundation
• Bias and variance of nearest neighbours, part 1 of 3Machine learning · Foundation
• One step of gradient descent, part 1 of 4Machine learning · Foundation
• Cross-validating a return forecaster, part 1 of 3Machine learning · Foundation
• Scoring a tree split, part 1 of 3Machine learning · Foundation
• Why is ordinary shuffled k-fold cross-validation wrong on a financial time series?Machine learning · Applied · Free solution
• Why do gradient-boosted trees usually beat deep networks on daily…Machine learning · Applied
• You cluster assets by return correlation with k-means and the result looks arbitrary.Machine learning · Applied
• A classifier assigns scores at random. What is its expected AUC?Machine learning · Applied
• In a ten-dimensional unit cube, what fraction of the volume lies within 0.1 of some face?Machine learning · Applied
• A support vector machine has |w|= 4. What is the width of its margin?Machine learning · Applied
• Gradient descent with momentum uses a learning rate of 0.01 and β = 0.9.Machine learning · Applied
• A batch normalisation layer follows a fully connected layer of 128 units.Machine learning · Applied
• A classifier flags 500 cases, of which 40 are true positives, and misses 60 positives.Machine learning · Applied
• Data are uniform in the unit hypercube in 10 dimensions.Machine learning · Applied
• A node has 50 positives and 50 negatives.Machine learning · Applied
• In gradient boosting, what happens if you halve the learning rate (shrinkage)…Machine learning · Applied
• Twenty percent of messages are spam.Machine learning · Applied
• What does the EM algorithm guarantee when fitting a Gaussian mixture model?Machine learning · Applied
• An anomaly detector has sensitivity 90% and specificity 95%.Machine learning · Applied
• You apply a strictly increasing transformation to a classifier’s scores.Machine learning · Applied
• Gradient descent on f(x) = x² starts at x = 1 with learning rate 0.1.Machine learning · Applied
• A quadratic loss has Hessian eigenvalues 4 and 0.5.Machine learning · Applied
• A convolutional layer uses 3 × 3 kernels, 16 input channels and 32 output…Machine learning · Applied
• Bias and variance of nearest neighbours, part 2 of 3Machine learning · Applied
• One step of gradient descent, part 2 of 4Machine learning · Applied
• Cross-validating a return forecaster, part 2 of 3Machine learning · Applied
• Scoring a tree split, part 2 of 3Machine learning · Applied
• What does embargoing do that purging does not?Machine learning · Advanced · Free solution
• A regularised linear model beats your tuned gradient-boosting ensemble out of sample.Machine learning · Advanced
• Two features are nearly identical, and a random forest reports low importance for both.Machine learning · Advanced
• What does the kernel trick actually buy you?Machine learning · Advanced
• You increase the batch size from 32 to 512 without changing anything else. What happens?Machine learning · Advanced
• Why did rectified linear units largely replace sigmoids in deep networks?Machine learning · Advanced
• You cluster a universe of assets from their correlation matrix.Machine learning · Advanced
• Each tree’s prediction has variance 1 and any two trees’ predictions correlate at 0.3.Machine learning · Advanced
• Adam’s first-moment estimate starts at 0 with β₁ = 0.9.Machine learning · Advanced
• Bias and variance of nearest neighbours, part 3 of 3Machine learning · Advanced
• One step of gradient descent, part 3 of 4Machine learning · Advanced
• Cross-validating a return forecaster, part 3 of 3Machine learning · Advanced