Sandboxes
No wrong answers here — slide the knobs and watch what the little network does.
What Can a Tiny Brain Decide?
Drag the glowing point around. The background shows what the little network thinks: blue means 'yes' and orange means 'no'. Move the sliders to change what it believes.
Imagine sorting animals: Class A = cats 🐱 (blue circles), Class B = dogs 🐶 (orange crosses). Every dot on the map is one example. The network draws a line that tries to separate cats from dogs — drag the point to see which side it lands on.
Drag the point on the canvas to test the network anywhere.
Control knobs
Activation Curve Playground
See how ReLU, Sigmoid and Tanh bend the same input. Slide the input dot along the curve and watch the output value change.
A neuron first multiplies its input by a weight and adds a bias. The activation function is the last step: it decides the shape of the 'boom'. Gain stretches the input before the curve sees it.
Control knobs
Make the Neurons Glow
A tiny 2-2-1 network. Push the inputs and bias around to light up the neurons. Brighter nodes are firing harder, and the connections pulse with their strength.
Control knobs
Signal Tracer — Follow the Numbers
Type inputs, click any connection to edit its weight, then watch the actual numbers flow: input → ×weight → sum + bias → activation → output.
This is the whole network with the lid off. Every line shows the signal travelling through it, every neuron shows its value, and you can rewire any weight or add hidden layers. Press Step to walk through one layer at a time.
Type the input values
Activation
Hidden layers
Connection weight
Click any line in the diagram to edit its weight.
Training Paradigms Lab
One target, three ways to learn it. Fit a line to five labelled points with supervised gradient steps, a reward-hungry agent, or a breeding population.
The orange dots are the training data. The network wants a line that passes through them. Switch between Supervised (learn from labels), Reinforcement (learn from reward) and Evolution (breed the fittest) — each uses a different knob, but they all chase the same shrinking error.
Control knobs
Gradient Descent Playground
Drop a ball on a loss curve and watch it roll downhill. Crank the learning rate and it overshoots the bottom; crank it further and it flies away.
The blue curve is the loss for every possible weight w: loss = w². The red ball takes gradient-descent steps. Start position and learning rate are yours to tune. The minimum sits at w = 0.
Control knobs