Foundations 0/0

Foundations

Advanced ML

Learning theory, uncertainty and conformal prediction, self-supervised representation learning, generative models such as VAEs and diffusion, and causal inference. These are the tools for defending a model’s confidence, its data efficiency or a causal claim made from its outputs.

0 / 27 solved

Questions

0/27 solved
Question Module Tags Difficulty Success
1. Four Times The Data Or A Quarter More Learning Theory & Phenomena
pac-learningvc-dimension +1
Medium —
2. The Late Bloomer Network Learning Theory & Phenomena
grokkinggeneralization +1
Hard —
3. Where The Curve Turns Twice Learning Theory & Phenomena
double-descentinterpolation-threshold +1
Medium 100%
4. When is Black Box Shift Estimation identifiable? Learning Theory & Phenomena
Machine LearningModels & Assumptions
Hard —
5. What occurs in the infinite-width NTK regime? Learning Theory & Phenomena
Machine LearningModels & Assumptions
Hard —
6. What does standard conformal prediction guarantee? Uncertainty & Bayesian
Machine LearningBayesian & Uncertainty
Medium —
7. Coverage Drifts Away Uncertainty & Bayesian
conformal-predictiondistribution-shift +1
Medium —
8. Doubt That Data Dissolves Uncertainty & Bayesian
uncertaintymc-dropout +1
Medium —
9. Why can deep ensembles improve uncertainty estimates? Uncertainty & Bayesian
Machine LearningBayesian & Uncertainty
Medium —
10. Which Calibration Score Becomes The Threshold Uncertainty & Bayesian
conformal-predictioncalibration +1
Medium —
11. What The Single Head Cannot See Uncertainty & Bayesian
deep-ensemblespredictive-uncertainty +1
Hard —
12. Averaging Out Your Doubts Uncertainty & Bayesian
bayesian-model-averagingprior-posterior +1
Hard —
13. One Layer Upstream Of The Loss Self-Supervised & Representation Learning
simclrcontrastive-learning +1
Hard —
14. Which VICReg term directly prevents complete collapse? Self-Supervised & Representation Learning
Machine LearningRepresentation Learning
Medium —
15. Why can an InfoNCE mutual-information bound saturate? Self-Supervised & Representation Learning
Machine LearningRepresentation Learning
Hard —
16. Stop, Collaborate, Don't Collapse Self-Supervised & Representation Learning
self-supervised-learningcontrastive-learning +1
Hard —
17. What characterises posterior collapse in a VAE? Generative Models
Machine Learning
Medium —
18. What Comes Out Of The Noise Schedule Generative Models
diffusionddpm +1
Medium —
19. The ELBO Went Up And The Model Got Worse Generative Models
vaeelbo +1
Medium —
20. Volume Costs Density Generative Models
normalizing-flowschange-of-variables +1
Medium —
21. Guided Past The Conditional Generative Models
diffusionclassifier-free-guidance +1
Medium —
22. What happens when causal overlap becomes weak? Causal Inference
Machine LearningCausal Inference & Policy Evaluation
Medium —
23. What is the main statistical role of cross-fitting? Causal Inference
Machine LearningCausal Inference & Policy Evaluation
Medium —
24. What does double robustness actually guarantee? Causal Inference
Machine LearningCausal Inference & Policy Evaluation
Medium —
25. Why can doubly robust off-policy evaluation outperform importance sampling? Causal Inference
Machine LearningCausal Inference & Policy Evaluation
Hard —
26. Blocking the Back Door Causal Inference
confoundersbackdoor adjustment
Hard —
27. What does Neyman orthogonality protect against? Causal Inference
Machine LearningCausal Inference & Policy Evaluation
Hard —