What does this Machine Learning cheat sheet cover?

It has 10 sections: Learning Setup, Linear Models, Optimization, Bias-Variance, Trees & Ensembles, SVMs, Clustering & PCA, Neural Nets, Evaluation and Validation. Together they cover the topics a typical Machine Learning exam tests, including supervised vs unsupervised, linear and logistic regression, loss functions and gradient descent and bias-variance tradeoff and regularization.

Is the machine learning cheat sheet free?

The first part is free to read on this page, with no account needed. The full sheet opens in the app with Unlimited ($4.99 a month or $19 for a semester), where you can edit it, print it or save it as a PDF. It is one US Letter page and prints cleanly in black and white or colour.

Can I bring this Machine Learning cheat sheet to my exam?

Only if your course allows a note sheet, so check the syllabus. This one is a single US Letter page. If your exam allows a different size, such as a 3x5 or 4x6 index card or one page front and back, Cheat Sheet AI makes one at exactly that size from your own lecture notes.

The sheet was written by our engine from a standard course outline, then a second model fact-checked every formula and statement against it; it corrected 1 statement before publishing. Notation and emphasis vary between courses, so check it against your own lecture notes.

Can I make a Machine Learning sheet from my own notes?

Yes. Upload your lecture slides, notes or study guide and Cheat Sheet AI writes a sheet in your course's notation, sized to exactly the space your exam allows. Your first sheets are free.

ComputingMachine Learning

Machine Learning Cheat Sheet — 1-page PDF

Supervised/unsupervised learning, core models, optimization, regularization, evaluation. The whole course on one dense, printable page for your midterm or final, with every formula typeset and worked examples where they help.

Make this from your notes Unlock the full sheet

The first part is free below. See pricing →

For a specific exam: Machine Learning Final exam study sheet.

1 pageUS Letter
10 sections10 topics
Free previewfull sheet with Unlimited

Preview of the one-page Machine Learning cheat sheet: Supervised/unsupervised learning, core models, optimization, regularization, evaluation — the whole course, one page

On the sheet

What’s on it

The topics a typical Machine Learning exam tests, all on the one page.

Supervised vs unsupervised
Linear and logistic regression
Loss functions and gradient descent
Bias-variance tradeoff and regularization
Decision trees and ensembles
SVMs
K-means and PCA
Neural networks and backprop
Evaluation metrics
Cross-validation

Free preview

The free part, as text

The free part of the sheet, section by section. Unlock the full sheet to read the rest, edit any block, and print it or save it as a PDF.

01Learning Setup

Supervised learning: Learn mapping $f : X \to Y$ from labeled data $(x_{i}, y_{i})$ ; tasks: regression ( $y \in R$ ), classification ( $y \in {1, \dots, K}$ ).
Unsupervised learning: Given only $x_{i}$ , find structure: clustering, dimensionality reduction, density estimation.

Common task types

Task	Input	Output
Regression	$x$	real $y$
Binary classification	$x$	$y \in {0, 1}$
Multiclass	$x$	$y \in {1.. K}$
Clustering	$x$	groups
PCA	$x$	low-dim $z$

Training / validation / test: Training fits parameters; validation tunes hyperparameters/model choice; test used once for final estimate.

Watch out:Data leakage: anything computed using test data (scaling, feature selection, PCA, CV folds) invalidates evaluation.

1. Split data or use CV.
2. Fit on training only.
3. Tune on validation.
4. Report final metric on untouched test set.

02Optimization

Mini-batch

θ \leftarrow θ - α \frac{1}{m} \sum_{j \in B} \nabla J_{j} (θ)

Most common practice.

Watch out:Too large $α$ ⇒ divergence; too small ⇒ slow convergence.

For convex losses: GD can reach global optimum.
Feature scaling often speeds GD.
Local minima/saddles matter for neural nets, less for linear models.

03Bias-Variance

Decomp.

E [(y - \hat{f} (x))^{2}] = Bias^{2} + Var + σ^{2}

Squared loss; $σ^{2}$ irreducible noise.

Bias: Error from overly simple assumptions; high bias ⇒ underfitting.
Variance: Sensitivity to training set; high variance ⇒ overfitting.

Model complexity effects

Effect	Bias	Var	Train err	Test err
Simpler	up	down	up	underfit
More complex	down	up	down	overfit

$L_{2}$ reg.

J (w) = \frac{1}{n} \sum_{i = 1}^{n} ℓ (\overset{y}{^}_{i}, y_{i}) + λ ∥ w ∥_{2}^{2}

Ridge / weight decay.

04SVMs

Hard-margin

min_{w, b} \frac{1}{2} ∥ w ∥^{2} s.t. y_{i} (w^{T} x_{i} + b) \geq 1

Linearly separable data; margin $= 2/ ∥ w ∥$ .

Soft-margin

min_{w, b, ξ} \frac{1}{2} ∥ w ∥^{2} + C \sum_{i} ξ_{i} s.t. y_{i} (w^{T} x_{i} + b) \geq 1 - ξ_{i}, ξ_{i} \geq 0

Allows violations; $C$ trades margin vs errors.

Support vectors: Training points with nonzero dual coefficients; lie on or inside margin and determine boundary.
Kernel trick: Replace dot products by $K (x, x^{'}) = ϕ (x)^{T} ϕ (x^{'})$ for nonlinear decision boundaries.

Common kernels: linear, polynomial, RBF $K (x, x^{'}) = e^{- γ ∥ x - x^{'} ∥^{2}}$ .

05Clustering & PCA

k-means: Partition $n$ points into $k$ clusters by minimizing within-cluster SSE.

k-means objective

min_{{C_{j}}} \sum_{j = 1}^{k} \sum_{x_{i} \in C_{j}} ∥ x_{i} - μ_{j} ∥^{2}

$μ_{j}$ = centroid of cluster $C_{j}$ .

06Neural Nets

Universal approximation: A sufficiently wide hidden-layer network can approximate many functions arbitrarily well (given suitable activation).

Watch out:Vanishing gradients: sigmoid/tanh in deep nets can make $\partial J / \partial W$ tiny; ReLU mitigates.

07Evaluation

Accuracy: $(T P + T N) / (T P + T N + F P + F N)$ ; best when classes balanced.

Precision

\frac{T P}{T P + F P}

Of predicted positives, fraction correct.

Recall

\frac{T P}{T P + F N}

Of actual positives, fraction found.

\frac{2 Precision \cdot Recall}{Precision + Recall}

Harmonic mean; balances P/R.

Specificity

\frac{T N}{T N + F P}

True negative rate.

ROC / AUC: ROC plots $T P R$ vs $F P R$ over thresholds; AUC summarizes ranking quality.

MSE / RMSE

MSE = \frac{1}{n} \sum_{i} (\overset{y}{^}_{i} - y_{i})^{2}, RMSE = MSE

Regression metrics.

Cheat sheet locked

Also on the full Machine Learning sheet

8 more sections, with every definition, formula and worked example. Unlimited unlocks every full sheet in the library to edit, print or save as a PDF, and makes unlimited sheets from your own notes.

Linear Models

Linear reg
MSE
Normal eqn
Logistic
Log loss
Decision rule

Optimization

Gradient descent
Gradient
Batch GD
SGD

Bias-Variance

L1 reg

Trees & Ensembles

Decision tree
Gini
Entropy
Pruning
Random forest
Boosting

SVMs

Maximum-margin classifier
Hyperplane

Clustering & PCA

PCA
PCA optimization
Covariance

Neural Nets

Neuron
Forward pass
Common activations
Backprop
Weight gradient

Validation

k-fold CV
Stratified CV
Train/val split
Metric choice

Unlock the full sheet $4.99 a month or $19 a semester

Or make this from your own notes

FAQ

Questions

About the Machine Learning sheet and making your own.

What does this Machine Learning cheat sheet cover?: It has 10 sections: Learning Setup, Linear Models, Optimization, Bias-Variance, Trees & Ensembles, SVMs, Clustering & PCA, Neural Nets, Evaluation and Validation. Together they cover the topics a typical Machine Learning exam tests, including supervised vs unsupervised, linear and logistic regression, loss functions and gradient descent and bias-variance tradeoff and regularization.
Is the machine learning cheat sheet free?: The first part is free to read on this page, with no account needed. The full sheet opens in the app with Unlimited ($4.99 a month or $19 for a semester), where you can edit it, print it or save it as a PDF. It is one US Letter page and prints cleanly in black and white or colour.
Can I bring this Machine Learning cheat sheet to my exam?: Only if your course allows a note sheet, so check the syllabus. This one is a single US Letter page. If your exam allows a different size, such as a 3x5 or 4x6 index card or one page front and back, Cheat Sheet AI makes one at exactly that size from your own lecture notes.
How accurate is it?: The sheet was written by our engine from a standard course outline, then a second model fact-checked every formula and statement against it; it corrected 1 statement before publishing. Notation and emphasis vary between courses, so check it against your own lecture notes.
Can I make a Machine Learning sheet from my own notes?: Yes. Upload your lecture slides, notes or study guide and Cheat Sheet AI writes a sheet in your course's notation, sized to exactly the space your exam allows. Your first sheets are free.

Keep going

All 40 sheets are in the sheet library.

Your course, your notes, one page.

Make this from your notes

Upload your own Machine Learning slides or notes and get a study sheet in your course’s notation. Your first sheet needs no account, and a free account adds one more. See pricing