Learning dashboard
Choose a mode, configure a layer, and solve the generated challenge.
Game setup
Shape DetectiveKeyboard: Enter checks answers, N creates a new challenge, and H reveals a hint.
Neural network arena
Visualise a dense layer and animate its forward pass.
Current challenge
Start the game to generate a question.
Configure the game, then press Start game.
Interactive workspace
Edit matrices, build layers, inspect calculations, and compare outputs.
Input matrix X
Weight matrix W
Bias vector b
Pre-activation Z
Activated output A
Architecture summary
- Start a challenge to generate a worked solution.
Session history
| Round | Mode | Result | Points | Time |
|---|---|---|---|---|
| No attempts yet. | ||||
Achievements
Core formulas
Concept guide
A dense layer connects every input feature to every neuron. Its weight matrix stores one weight for every connection.
The batch dimension remains unchanged. The units setting determines the final dimension of the layer output.
Bias adds one trainable value per neuron. Disabling bias removes those values from the parameter total.
Activation functions transform weighted sums. Softmax converts each output row into probabilities that sum to one.
Common errors
| Error | Cause | Fix |
|---|---|---|
| Input mismatch | Input columns differ from weight rows. | Set weight rows equal to input features. |
| Bias mismatch | Bias length differs from output units. | Use one bias value per neuron. |
| Layer mismatch | A layer's units differ from the next layer's inputs. | Connect each layer using the previous output size. |
| Wrong softmax axis | Softmax is applied across samples. | Apply softmax across each output row. |
Custom challenges and exports
Save configurations, share results, and continue learning later.
Generated on 2026-08-10 15:36:50. Progress and custom challenges remain in this browser unless cleared.