Microsoft ML & AI interview questions.
Questions in the CrackML corpus associated with Microsoft. Use company evidence as a prioritization signal, not a promise that an exact question will repeat.
How is BERT different from GPT?GenAI & LLM · medium · Evidence 88/100Explain Transformer architecture and each major component.GenAI & LLM · medium · Evidence 88/100Explain the bias-variance trade-off.ML Fundamentals · easy · Evidence 77/100Solve the maximum-points-on-a-line problem.Python & DSA · hard · Evidence 77/100Compare covariate shift and label drift.ML Fundamentals · medium · Evidence 77/100Implement self-attention.ML Coding & PyTorch · hard · Evidence 77/100What are the trade-offs of downsampling the majority class?ML Fundamentals · medium · Evidence 77/100Compare Thompson sampling and epsilon-greedy.ML Fundamentals · hard · Evidence 77/100How would you handle cold start for new ads?ML System Design · medium · Evidence 77/100Implement masked self-attention.ML Coding & PyTorch · hard · Evidence 77/100Why can offline metrics look better than online metrics?ML Fundamentals · hard · Evidence 77/100Explain the exploration-exploitation trade-off.ML Fundamentals · medium · Evidence 77/100Solve merge intervals.Python & DSA · medium · Evidence 77/100Which statistical tests can detect distribution drift?ML Fundamentals · hard · Evidence 77/100How would you vectorize a K-means implementation?ML Coding & PyTorch · hard · Evidence 77/100How would you learn a new policy from logged data collected by an old policy?ML Fundamentals · hard · Evidence 77/100Implement K-means from scratch.ML Coding & PyTorch · medium · Evidence 77/100How do you calibrate a classifier?ML Fundamentals · medium · Evidence 75/100Implement self-attention from scratch in code.ML Coding & PyTorch · hard · Evidence 75/100Analyze the space complexity of major Transformer components.GenAI & LLM · hard · Evidence 75/100Design a RAG system that generates presentation slides from a user prompt.GenAI & LLM · hard · Evidence 75/100Explain how RAG works end to end.GenAI & LLM · medium · Evidence 75/100Compare bagging and boosting.ML Fundamentals · medium · Evidence 58/100How would you investigate model drift in production?ML System Design · medium · Evidence 58/100What happens to logistic regression on perfectly separable data?ML Fundamentals · medium · Evidence 31/100