Home▸AI & Machine Learning▸Neural Collaborative Filtering in 2D: Watch Embeddings Learn Live

Neural Collaborative Filtering in 2D: Watch Embeddings Learn Live

A real Neural Collaborative Filtering model trained live on a 2D canvas: watch user and movie embeddings drift across a pannable, zoomable scatter plot while a node-link diagram and loss/accuracy sparkline show backpropagation through a small MLP replacing the plain dot product of matrix factorization.

AI & Machine Learning2DAdvanced60 FPS📱 Mobile-adapted⇄ 3D version

Details

CategoryAI & Machine Learning

RenderingCanvas 2D

DifficultyAdvanced

Target FPS60

ControlsSliders for hidden-layer width and learning rate, Train/Pause and single-Step buttons to run gradient descent, a Reset button to reinitialize embeddings and weights, a viewer dropdown that draws live recommendation-score lines on the 2D scatter plot, and drag-to-pan / scroll-to-zoom on the embedding panel itself.

recommender-systemsneural-networkscollaborative-filteringbackpropagationembeddingsmachine-learning

Related simulations

How Recommendation Engines Decide What You SeeInteractive 3D user-item embedding space where adjusting a simulated user's interaction…

3D Neural Network XOR Training SimulationWitness a 3D neural network learn the XOR problem in real-time by adjusting learning rate,…

Rotation Prediction: A Self-Supervised Pretext TaskWatch an embedding space self-organize as a rotation-prediction pretext task trains:… ⊞ All 1161 in AI & Machine Learning →

Most recommender demos stop at a dot product. This one trains a real, small Neural Collaborative Filtering model in your browser — five viewers and six movies, each embedded as a point on a 2D scatter plot, feeding into a hidden layer that learns a non-linear scoring function via backpropagation. Press Train and watch full-batch gradient descent reshape the embedding space epoch by epoch: viewers who share taste drift toward each other, movies drift toward the viewers who'd enjoy them, a node-link diagram shows the weights themselves changing sign and strength, and a live sparkline tracks loss and accuracy converging. Pick a viewer to see the model's current recommendation scores drawn as live connecting lines, and drag or scroll on the scatter plot to explore the space up close.

⚙ Under the hood

Train a real Neural Collaborative Filtering model live on a 2D canvas: watch user and movie embeddings drift across a pannable, zoomable scatter plot while a node-link diagram and a loss/accuracy sparkline show backpropagation through a small MLP that learns a non-linear interaction function, replacing the plain dot product of classic matrix factorization.

recommender-systemsneural-networkscollaborative-filteringbackpropagationembeddingsmachine-learning

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install