Credits - ML Resources Hub

How to read this page. The table lists each topic's most direct inspiration — the work whose layout, interaction, or visual idiom I leaned on. The bottom section lists broader influences whose aesthetic shows up across the site.

Per-topic inspirations

For each topic, the source whose visual pattern is most visibly present in our viz.

Viz	Closest prior art
CNN	CNN Explainer (Wang et al., Georgia Tech) — kernel cells, sliding receptive field, mini-heatmap output panel.
Transformer	BertViz (Vig) for arcs-over-tokens; Transformer Explainer (Cho et al.) for layout; The Illustrated Transformer (Alammar) for the head-tab presentation.
RNN	Understanding LSTM Networks (Olah) — the unrolled-timesteps diagram with shared cell is essentially his.
GNN	A Gentle Introduction to Graph Neural Networks (Sanchez-Lengeling et al., Distill) — click-a-node propagation is straight from theirs.
Autoencoder / VAE	Influenced by interactive latent-walk demos in Lilian Weng's tour of VAEs and the broader genre of "decoder slider" toys that's accumulated across many blog posts.
GAN	GAN Lab (Kahng, Thorat, Chau, Wei) — the particle generator vs. discriminator decision-surface + loss curves layout is essentially GAN Lab simplified.
Diffusion	The forward formula x_t = √α̅·x₀ + √(1-α̅)·ε is from DDPM (Ho, Jain, Abbeel). The cosine schedule is from Improved Denoising Diffusion (Nichol & Dhariwal). The thumbnail-strip-over-time visual style is common across diffusion blog posts.
DNN	Aesthetic borrows from TensorFlow Playground (Smilkov & Carter) — small-network forward pass with per-neuron heatmaps.
Regression	Setosa.io — OLS Regression (Powell & Lehe) — drag-points-watch-line interaction.
k-means	Visualizing K-Means Clustering (Harris) — the step-through-Lloyd's interaction with the assignment / update split.
Decision Trees	R2D3 — A Visual Introduction to Machine Learning (Yee & Chu) and scikit-learn's decision-boundary plot examples.
Random Forests	R2D3 Part 1's tree-ensemble illustrations, plus scikit-learn's forest-vs-tree comparison plots.
SVM	Setosa.io's approach to interactive 2D classifier viz; scikit-learn's SVM example plots for the linear-margin layout.
Gradient Boosting	Various blog posts walking through residual-fitting visually; the data + target + fit + residuals layout is in explained.ai's gradient boosting tutorial (Parr & Howard).
GMM	scikit-learn's GMM ellipse plots set the canonical visual idiom — full-covariance components drawn as 1σ ellipses with points coloured by responsibility.
Gaussian Processes	A Visual Exploration of Gaussian Processes (Görtler, Kehlbeck, Deussen — Distill) — the canonical "click to add observations, watch the band collapse, draw samples from the posterior" interaction.
HMM	The activity-as-emoji + state-posterior-as-heatmap presentation is influenced by Stanford CS228's lecture notes on HMMs; the Viterbi-as-line overlay is a long-standing textbook idiom going back to Rabiner's 1989 tutorial.
Naive Bayes	scikit-learn's classifier-comparison gallery; per-class diagonal Gaussian + axis-aligned ellipse + class-mix heatmap is its standard visual.
Dim Reduction (PCA)	Setosa.io — Principal Component Analysis (Powell & Lehe) — the "rotate the projection axis, watch the histogram widen, find the variance-maximising angle" interaction is theirs.
Bayesian Inference	Setosa.io — Conditional Probability for interaction style. The Beta-Binomial coin-flip demonstration is in every Bayesian textbook (Sivia, BDA3, Kruschke), going back to classical conjugate-prior worked examples.
Anomaly Detection	scikit-learn's outlier-detection example set the colour conventions (low-density background + flagged points with markers).
Bias-Variance	The "many fits on resampled training sets + U-curve of bias² and variance against complexity" is Figure 7.1 of The Elements of Statistical Learning (Hastie, Tibshirani & Friedman) made interactive.
Gradient Descent	Why Momentum Really Works (Goh, Distill) for the loss-surface aesthetic; TensorFlow Playground for the live-trace style.

Broader influences

Sources whose aesthetic, vocabulary, or pedagogical approach shows up across many of the visualisations rather than being tied to one topic.

Distill.pub

The journal that set the modern bar for interactive ML explanation. Their type, palette, and "exploration as scrollytelling" influence is everywhere on this site.
Polo Club of Data Science

Georgia Tech's visualisation group. CNN Explainer, GAN Lab, Transformer Explainer, Diffusion Explainer — all directly shaped how I built mine.
Setosa.io / Explained Visually

Powell & Lehe's collection of statistical visualisations. The clean, single-purpose, "play with one slider" aesthetic permeates this site's classical-ML viz.
Christopher Olah

Pioneered the modern "deep learning concept as gentle visual essay" form. The RNN viz here is almost a direct homage.
Jay Alammar

The Illustrated Transformer / BERT / GPT — the visual vocabulary for explaining attention layered into how the transformer viz here is presented.
R2D3

Yee & Chu's interactive ML intro. Set the standard for decision-boundary scrollytelling in classical ML.
scikit-learn examples

The baseline visual reference for nearly every classical algorithm here — particularly for decision boundaries, GMM ellipses, and classifier comparisons.
Elements of Statistical Learning

Hastie, Tibshirani & Friedman. Many of the static figures in ESL anchor how we picture these methods; the bias-variance viz is the most direct lift.

If you're an author of one of the works above and want a different attribution (or want yours removed), email me at liv.helen.vage@cern.ch. The intent here is to make the lineage visible, not to claim credit.

The code for the visualisations on this site is original implementation in vanilla JavaScript / Canvas, but it would be dishonest to pretend the ideas are. They aren't.