The 15-Day AI/ML Interview Bootcamp
Day 5 — Overfitting, the bias–variance trade-off, and regularisation
Why a model that aces training data fails in production, the bias-variance trade-off drawn rather than stated, L1 and L2 regularisation from scratch, k-fold cross-validation, learning curves, and how to tell 'more data' from 'simpler model' — the single most asked machine learning interview topic.
It's late, and the number on your screen is beautiful: training loss 0.0753. You push the model, go to bed, and on Monday the first real batch comes back wrong on a third of the rows. Nothing crashed. The code is the same code. The model you trained simply learned your twenty rows — the noise in them included — instead of the thing that made them. Today you and I will make that failure happen on purpose, name the two ways a model can be wrong, and build the three tools that fix it: a leash on the weights, a leash that cuts, and a way of scoring a model that can't be fooled. Yesterday's assignm…
In this lesson
Unlock the rest of The 15-Day AI/ML Interview Bootcamp
Day 5 — Overfitting, the bias–variance trade-off, and regularisation is part of the full course. One payment unlocks every lesson, every other course, and every premium article — for life.