Home▸AI & Machine Learning▸Autoencoder Reconstruction Error Anomaly Detection Lab

🧠 Autoencoder Reconstruction Error Anomaly Detection Lab

A 3D lab showing how an autoencoder learns a 'normal' data manifold and flags anomalies by reconstruction error, with a bottleneck-capacity control and a density-based comparison mode.

AI & Machine Learning3DAdvanced60 FPS

An autoencoder trained only on normal data learns a smooth "normal" manifold; every point's distance from that manifold is its reconstruction error, and points scoring above a threshold get flagged as anomalies.

🔬 What It Demonstrates

The purple surface is the manifold the autoencoder has learned to reconstruct faithfully. Green points sit close to it (low error, normal); red points sit far away (high error, flagged). Raising latent bottleneck capacity too far lets the network start reconstructing anomalies too, quietly eroding detection power.

🎮 How to Use

Adjust anomaly rate, bottleneck size and the error threshold, and watch flagged/caught/false-alarm counts update live. Switch to density-based scoring to compare against nearest-neighbour methods like DBSCAN, LOF and One-Class SVM.

💡 Did You Know?

Reconstruction-error autoencoders need no labelled anomalies to train — just a clean sample of normal behaviour — which is why they're a go-to for fraud, sensor-fault and intrusion detection where true anomalies are rare and always changing.

⚙ Under the hood

A 3D lab showing how an autoencoder learns a 'normal' data manifold and flags anomalies by reconstruction error, with a bottleneck-capacity control and a density-based comparison mode.

autoencoderanomaly detectionreconstruction errorneural networksmachine learningdeep learningThree.js

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install