ClearDataLabs
Articles
-
The Architectural Loophole: AI Copyright & Book Replication
Discover how AI refuses copyrighted PDFs but recreates entire book structures from memory. Explore the ethics of Architectural Replication and synthesized displacement.
-
What ClearDataLabs Is About
ClearDataLabs is an open-source project that explains AI and neural networks through interactive browser demos and in-depth articles — building everything from scratch, no ML libraries.
-
Seeing in Cells: How Handwriting Becomes Input for a Neural Network
How to preprocess handwriting for a neural network: raw pen strokes become a scale-invariant 12×12 binary grid of 144 numbers — implemented in TypeScript, no libraries.
-
Making a Neural Network Dream: DeepDream-Style Visualization in JavaScript
A DeepDream-style technique in pure JavaScript: optimize a blank image until a neural network confidently sees a digit. Visualize what an MNIST network has learned.
-
What Does a Neural Network Learn? Visualizing MNIST with Causal Index
What has a neural network trained on MNIST actually learned? We compute a causal index per pixel and visualize the result as a heat map — implemented in pure JavaScript.
-
Handwriting Recognition in the Browser — Neural Network in TypeScript from Scratch
A handwriting recognition neural network running entirely in the browser, built in TypeScript with no ML libraries. Train it on your own letters and watch it classify in real time.
-
Backpropagation in the Browser: Training a Neural Network in JavaScript
How backpropagation and gradient descent work, step by step — implemented in TypeScript and running live in the browser. Train a neural network without leaving JavaScript.
-
Building a Neural Network from Scratch in TypeScript — No Libraries
How to build a feedforward neural network from scratch in TypeScript: neurons, weights, sigmoid activation, and forward pass — a 3-layer perceptron in under 200 lines, no ML libraries.