The bottom of the stack. Nothing above this makes sense without it. Titles are on the spines; the notes underneath are the three tellings, in short — for a ten-year-old, for the architect, and for the boardroom.
6 concepts on this shelf · Beginner → Intermediate · about 2 min read
Kitchen table — Instead of writing down every rule, you show it thousands of examples and let it work out the rule itself.
Whiteboard — Supervised, unsupervised, and everything in between: fit a function to labelled data, then hope the distribution at inference looks like the distribution you trained on.
Boardroom — It learns your patterns from your data — which means the quality of the answer is capped by the quality of the history you feed it.
machine learning, supervised learning, unsupervised learning, training data, model fitting
Kitchen table — Lots of tiny switches in rows. Each one passes a little of the signal on, and together they notice things no single switch could.
Whiteboard — Layers of weighted sums and non-linearities. Depth buys representation; width buys capacity; backpropagation is how the blame gets shared.
Boardroom — The engine behind almost everything labelled AI today. Powerful, expensive to train, cheap to run, and hard to interrogate after the fact.
neural network, deep learning, layers, activation function, backpropagation
Kitchen table — You are on a foggy hill and you want to reach the bottom. You cannot see far, so you feel which way is downhill and take one step. Then again.
Whiteboard — Differentiate the loss with respect to the parameters, step against the gradient, argue about the learning rate for the rest of your career.
Boardroom — This is literally what 'training' means — millions of small corrections. It is why training takes time and money and why it eventually stops improving.
gradient descent, optimization, learning rate, slope, derivative, SGD, Adam
Kitchen table — A score for how wrong the guess was. The whole game is making that number smaller.
Whiteboard — MSE for regression, cross-entropy for classification, and the quiet truth that your loss function is the only definition of 'good' the model will ever see.
Boardroom — Whatever you choose to measure is what the system will optimise for — including the things you did not mean to reward.
loss function, mean squared error, MSE, cross entropy, objective function, error
Kitchen table — Computers only do numbers. So 'Math', 'Science' and 'Art' have to become numbers first — without pretending Art is bigger than Math.
Whiteboard — One-hot, ordinal, target and hashing encoders; multi-label columns; leakage; and the reminder that feature work still beats model choice more often than anyone admits.
Boardroom — Most of the effort in a real AI project is not the model. It is getting messy business data into a shape a model can use.
feature engineering, one-hot encoding, categorical variables, ordinal encoding, multi-label, preprocessing
Kitchen table — Before you trust it, you give it questions you already know the answers to, and count how many it gets right.
Whiteboard — Hold-out and cross-validation, precision against recall, class imbalance, calibration, and why a single accuracy number is usually hiding something.
Boardroom — Ask for the metric and the baseline together. '92% accurate' means nothing until you know what guessing would have scored.
model evaluation, accuracy, precision, recall, F1, ROC AUC, cross validation, baseline