Home▸Biology & Life▸2D Neural Network — Backpropagation on a Flat Canvas

2D Neural Network — Backpropagation on a Flat Canvas

A flat 2D neural network trainer: real matrix forward passes, real chain-rule backpropagation and real gradient descent on XOR, circle or spiral data, drawn as a plain top-down diagram with weight lines, node activations and a live loss curve — no camera, no WebGL.

Biology & Life2DModerate60 FPS📱 Mobile-adapted⇄ 3D version

Details

CategoryBiology & Life

RenderingCanvas 2D

DifficultyModerate

Target FPS60

ControlsA task selector picks the dataset (XOR, circle boundary or two-class spiral). Sliders set hidden-layer width (2-16 neurons), learning rate (0.01-0.20) and training speed (epochs per animation frame). An activation selector switches the hidden layer between sigmoid and ReLU. Train/Pause toggles continuous training, Step advances one epoch, and Reset weights reinitializes the network. Clicking inside the decision-boundary panel feeds that point through the network and updates the diagram with its prediction.

neural networkbackpropagationgradient descentmachine learningxor problemsigmoid2d canvas

Related simulations

Game of LifeConway's cellular automaton in 3D: Conway, HighLife, Day&Night modes. Flat / height-map /…

Aquarium — Fish Schooling SimulationExplore the dynamic behavior of fish as they school together in this realistic aquarium…

SIR Epidemic ModelParticles spread infection by proximity. Tune β, γ, vaccinate and watch R₀ and the epidemic… ⊞ All 153 in Biology & Life →

This is the flat, top-down companion to the 3D Neural Network — Backpropagation simulator. It runs the exact same real matrix forward pass, chain-rule backpropagation and gradient descent on XOR, circle or spiral data, but draws the network and decision boundary as plain flat 2D canvas shapes — no camera, no WebGL, no rotation.

⚙ Under the hood

neural networkbackpropagationgradient descentmachine learningxor problemsigmoid2d canvas

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