Skip to content
Scrollchart
🤖

Free Animated AI & Machine Learning Diagrams

Neural nets, transformers, training, classical ML, RAG, agents, evaluation. The shapes of modern AI.

80 live For: AI/ML bloggers · Research summarizers · ML educators · Applied practitioners

For AI/ML bloggers, research summarizers, and practitioners. Gradient descent on a loss surface, transformer attention flow, K-means clustering, PCA projection, gradient flow through layers, RAG retrieval pipelines, agent loops, scaling-law curves, eval-metric trade-offs.

What this is
Free, embeddable, animated diagrams covering ai & machine learning topics (neural networks, transformers, training, classical ml, and more).
Who uses it
AI/ML bloggers · Research summarizers · ML educators · Applied practitioners
How to embed
Copy one line of HTML per diagram. No signup. No watermark. Works in WordPress, Webflow, Ghost, Substack, plain HTML.
Diagrams in this category
80 live (total in catalog: 80).
License
Free forever. The diagram and editorial explainer text are served from scrollchart.com so they can be updated over time to reflect new research.
URL
scrollchart.com/categories/ai-ml
Highlights

What you will find here

  • Neural nets: forward/backprop, activations, gradient flow
  • Transformers: attention, position, MoE, KV cache
  • Training: optimizers, learning-rate schedules, regularization
  • Classical ML: K-means, PCA, decision trees, SVM
  • RAG, agents, evaluation, scaling laws
Sub-categories

6 topic areas in AI & Machine Learning

Neural networks

Forward/backprop, activations, gradient flow, vanishing/exploding.

16 live
Transformers

Attention, multi-head, position, MoE, KV cache, MLA.

16 live
Training

SGD, Adam, LR schedules, regularization, gradient clipping.

12 live
Classical ML

K-means, PCA, decision trees, SVM, naive Bayes, logistic regression.

14 live
RAG & agents

RAG pipelines, embedding spaces, agent loops, tool use, planning.

10 live
Evaluation & scaling

ROC, confusion matrix, scaling laws, perplexity, eval suites.

12 live
Catalog

Browse AI & Machine Learning diagrams

All categories →

Neural networks 16 live

Forward/backprop, activations, gradient flow, vanishing/exploding.

AI & Machine Learning Rich

Forward Pass: Inputs to Logits

Activations flow layer by layer. Linear, then nonlinearity, then linear again, ending in logits.
AI & Machine Learning Rich

Backpropagation: Gradients Flow Backward

Loss at the end. Gradients propagate backward via the chain rule, updating each weight.
AI & Machine Learning Rich

Activation Functions Compared

ReLU, sigmoid, tanh, GELU, SiLU. Curves, gradients, and where each shines.
AI & Machine Learning Rich

Gradient Flow vs Network Depth

In deep networks gradients shrink or explode unless architecture cooperates. Depth is hard.
AI & Machine Learning Medium

Vanishing & Exploding Gradients

Repeated multiplication by small numbers vanishes. By large numbers, explodes. The depth tax.
AI & Machine Learning Medium

Batch Normalization

Normalize activations within each minibatch. Steadier training, faster convergence.
AI & Machine Learning Medium

Layer Normalization

Normalize per example over the feature axis. Batch-size-agnostic, transformer default.
AI & Machine Learning Medium

Dropout: Stochastic Regularization

Randomly zero a fraction of activations during training. Forces redundancy, prevents overfitting.
AI & Machine Learning Medium

Residual Connections

A shortcut adds the input to the output. Gradients have a highway; networks scale to hundreds of layers.
AI & Machine Learning Medium

Convolution: Kernel Sliding Across Image

A small kernel sweeps the image computing dot products. Spatial features emerge.
AI & Machine Learning Simple

Pooling: Max & Average

Pool 2x2 to one value. Spatial downsample, translation tolerance.
AI & Machine Learning Rich

RNN vs LSTM

Vanilla RNN forgets fast. LSTM gates carry information across long sequences.
AI & Machine Learning Simple

Embedding Lookup

Discrete IDs index into a learned vector table. Words, items, and users become dense vectors.
AI & Machine Learning Medium

Softmax + Cross-Entropy

Logits become probabilities, target class loss is -log(p_correct). The classification staple.
AI & Machine Learning Simple

Overfit vs Underfit

Training loss falls; validation loss bends back up at overfit. The classic two-curve picture.
AI & Machine Learning Medium

Parameter Count vs Compute

Params, layers, and FLOPs scale together. Where AlexNet, ResNet, GPT-3, GPT-4 sit on the curve.

Transformers 16 live

Attention, multi-head, position, MoE, KV cache, MLA.

AI & Machine Learning Rich

Multi-Head Attention

Each head learns its own Q/K/V projections. Many heads, many relational lenses, concatenated.
AI & Machine Learning Rich

Transformer Block

Attention, residual, norm, MLP, residual, norm. The unit that stacks 100+ deep.
AI & Machine Learning Rich

Positional Encoding & RoPE

Attention is order-blind. Add positional signal: sinusoidal, learned, or rotary.
AI & Machine Learning Rich

KV Cache: Why Generation Speeds Up

During generation, past K and V are cached. Each new token only computes its own.
AI & Machine Learning Medium

Self vs Cross Attention

Self: Q, K, V from same sequence. Cross: Q from one, K/V from another. Decoder bridge.
AI & Machine Learning Medium

Causal Attention Mask

Upper-triangular mask prevents tokens from peeking forward. Required for autoregressive training.
AI & Machine Learning Rich

Mixture of Experts (MoE)

A router picks top-k experts per token. Sparse compute, dense capacity.
AI & Machine Learning Rich

FlashAttention: Tiled & Fused

Compute attention in blocks that fit in SRAM. Fused, online-softmax, no materialized N x N.
AI & Machine Learning Rich

MLA: Multi-Head Latent Attention

Compress K and V to a low-rank latent before caching. Massive cache shrink, near-MHA quality.
AI & Machine Learning Medium

GQA & MQA: Sharing K/V Heads

MHA: heads each have K and V. MQA: one shared K/V. GQA: groups share. Quality vs cache size.
AI & Machine Learning Rich

Long Context: Sliding Window & Sink

Local windows + a few global tokens preserve quality at 100k+ context.
AI & Machine Learning Medium

BPE Tokenization

Greedy merges build a vocab from bytes. The same word can become 1 or 5 tokens.
AI & Machine Learning Medium

Encoder-Decoder vs Decoder-Only

Original Transformer split. GPT keeps the decoder. T5 keeps both. Different fits, same primitives.
AI & Machine Learning Medium

Decoder-Only Scaling

GPT-2, GPT-3, GPT-4, Llama series. Same recipe scaled in width, depth, and data.
AI & Machine Learning Medium

Diffusion vs Autoregressive Generation

AR generates tokens left to right. Diffusion denoises a whole sample over many steps.
AI & Machine Learning Rich

Speculative Decoding

A small draft model proposes tokens; the big model verifies in parallel. 2-3x throughput.

Classical ML 14 live

K-means, PCA, decision trees, SVM, naive Bayes, logistic regression.

AI & Machine Learning Rich

K-Means Clustering Iterations

Assign each point to its nearest centroid. Move centroids to cluster mean. Repeat.
AI & Machine Learning Rich

PCA: Principal Components

Find directions of greatest variance. Project to fewer dimensions, keep most signal.
AI & Machine Learning Rich

Decision Tree: Gini Splits

At each node, find the split that most reduces impurity. Recursive partitioning of feature space.
AI & Machine Learning Rich

SVM: Maximum Margin Hyperplane

Find the hyperplane that maximizes margin to the nearest points. Kernels for nonlinear.
AI & Machine Learning Medium

Logistic Regression

Linear logits squashed by sigmoid. Probabilities for binary classification.
AI & Machine Learning Medium

Naive Bayes

Bayes' rule with independence assumption. Surprisingly strong on text and spam.
AI & Machine Learning Medium

Random Forest

Many decision trees voting, each trained on a bootstrap sample with feature subsampling.
AI & Machine Learning Rich

Gradient Boosting

Each new tree fits the residuals of the previous. Sequential, not parallel like RF.
AI & Machine Learning Simple

k-Nearest Neighbors

Look at the k closest training points and vote. Simple, lazy, surprisingly competitive.
AI & Machine Learning Rich

Gaussian Mixture & EM

Soft clustering: each point belongs partially to each Gaussian. EM alternates expectation and maximization.
AI & Machine Learning Medium

Hierarchical Clustering Dendrogram

Agglomerative bottom-up merges. Cut the tree at any height to choose cluster count.
AI & Machine Learning Rich

t-SNE vs UMAP

Two ways to embed high-D data into 2D. Different math, different distortions.
AI & Machine Learning Simple

Bias-Variance Tradeoff

Underfit: high bias, low variance. Overfit: low bias, high variance. The fundamental dial.
AI & Machine Learning Medium

Feature Engineering Pipeline

Raw data -> imputation -> encoding -> scaling -> interaction terms. Old school, still wins on tabular.

Training 12 live

SGD, Adam, LR schedules, regularization, gradient clipping.

AI & Machine Learning Rich

SGD vs Adam vs Lion

Three optimizers walking the same loss landscape. Different paths, different memory cost.
AI & Machine Learning Medium

LR Schedules: Warmup, Cosine, Cyclical

Linear warmup, then decay. Cosine and cyclical are the modern defaults.
AI & Machine Learning Rich

Mixed Precision Training

Compute in fp16/bf16, accumulate in fp32. Half the memory, near-full quality.
AI & Machine Learning Medium

Gradient Clipping

When gradient norm exceeds threshold, scale it down. Stops a single bad batch from destroying training.
AI & Machine Learning Simple

Weight Decay & L2 Regularization

Penalize large weights. Pulls solutions toward simpler, more generalizable models.
AI & Machine Learning Simple

Early Stopping

Watch validation loss. Stop when it starts climbing back up.
AI & Machine Learning Medium

Knowledge Distillation

A small student learns from a big teacher's soft logits. Compression without crashing accuracy.
AI & Machine Learning Rich

RLHF / RLAIF Pipeline

SFT -> reward model -> PPO/DPO. Three stages that align an LLM to preferences.
AI & Machine Learning Rich

Data, Tensor, and Pipeline Parallelism

Three axes of distribution. Each splits something different across GPUs.
AI & Machine Learning Medium

Batch Size Scaling Laws

Bigger batches enable bigger LR. Critical batch size beyond which returns vanish.
AI & Machine Learning Medium

Curriculum Learning

Start easy, get hard. Ordered data improves convergence on tough tasks.
AI & Machine Learning Rich

Loss Landscape Visualization

A slice through high-D loss reveals saddles, ridges, and basins. Why optimization is hard.

Evaluation & scaling 12 live

ROC, confusion matrix, scaling laws, perplexity, eval suites.

AI & Machine Learning Rich

ROC Curve & AUC

TPR vs FPR as the threshold sweeps. AUC summarizes the whole curve in one number.
AI & Machine Learning Rich

Confusion Matrix Anatomy

TP, FP, FN, TN. Every classifier metric falls out of these four counts.
AI & Machine Learning Rich

Scaling Laws: Chinchilla vs Kaplan

Loss as a power law in params, data, compute. Chinchilla redrew the optimal frontier.
AI & Machine Learning Medium

Perplexity vs Parameters

Bigger model, lower perplexity. Diminishing returns, but consistent.
AI & Machine Learning Medium

Precision vs Recall

Precision: of predicted positives, how many right. Recall: of actual positives, how many caught.
AI & Machine Learning Medium

Calibration: Are Probabilities Honest?

A 70%-confident model should be right 70% of the time. The reliability diagram shows it.
AI & Machine Learning Simple

k-Fold Cross-Validation

Split into k folds. Train on k-1, evaluate on one. Rotate. Average the score.
AI & Machine Learning Medium

LLM Eval Suites Compared

MMLU, GSM8K, HumanEval, MATH, BBH, ARC. What each measures and where they saturate.
AI & Machine Learning Medium

Leaderboard Overfit

Models tuned on a benchmark. Held-out variants reveal the gap.
AI & Machine Learning Medium

Pretraining Loss Curve

Smooth power-law decline over hundreds of billions of tokens. The shape of LLM pretraining.
AI & Machine Learning Medium

Human Preference Elo

Pairwise human votes feed an Elo system. Models rank by who wins more.
AI & Machine Learning Medium

Data Quality vs Quantity

Curated, deduped, mixed-domain data trains stronger models than larger raw dumps.
FAQ

Common questions about AI & Machine Learning diagrams

Where can I get free animated ai & machine learning diagrams for my website?

Scrollchart is a free library of animated, embeddable diagrams. The AI & Machine Learning section covers the most common concepts and frameworks in the field. Each diagram embeds with one line of HTML, no account required, no watermark.

How do I embed a free animated ai & machine learning diagram in WordPress?

Copy the embed snippet from the diagram page on Scrollchart and paste it into a custom HTML block in your WordPress editor. The diagram loads automatically and scroll-paints into view. No plugin or API key needed.

Is there a free library of animated ai & machine learning diagrams I can embed without coding?

Scrollchart provides ready-to-embed animated diagrams across ai & machine learning and 35 other categories. The embed is a single line of HTML that works in any CMS, including WordPress, Webflow, Ghost, Substack, and plain HTML pages.

Do Scrollchart ai & machine learning diagrams include explanatory text?

Yes. Each diagram ships with a tabbed explainer panel covering what the diagram shows, the research behind it, and how to interpret it. The text is served from scrollchart.com and rendered directly in your page's DOM so it is indexed by search engines.

Don't see what you need?

Generate a custom ai & machine learning diagram from a prompt, or request a new template. Most ship within two weeks.

Try the AI generator Request a diagram