ML Fundamentals Breadth Interview

Prompt Breadth screen across ML fundamentals and applied NLP: bias–variance, probability, regularization, RNNs, attention, seq2seq.

Be ready to discuss

• Bias–variance tradeoff: decomposition of generalization error, diagnosing under/overfitting from train-vs-val curves, and the knobs on each side (model capacity, data, regularization).
• Probability basics: Bayes rule, conditional independence, expectation/variance, common distributions, MLE vs MAP.
• Regularization: L1 (sparsity) vs L2 (weight shrinkage), dropout, early stopping, data augmentation, batch norm's regularizing side-effect.
• RNNs: recurrence, vanishing/exploding gradients, LSTM/GRU gating as the fix, truncated BPTT.
• Attention and seq2seq: encoder–decoder bottleneck, additive vs dot-product attention, how attention removes the fixed-length context constraint; transformers as the follow-on.
• Evaluation instincts: train/val/test discipline, metrics beyond accuracy (precision/recall, AUC, perplexity for LMs).