TensorFlow Playground

Question 1

TensorFlow Playground lets you train a small neural network…

Accepted Answer

Interactively in the browser, watching the decision boundary form in real time — It's a visual, no-code sandbox for building intuition about how networks learn to separate data.

Question 2

Adding more HIDDEN NEURONS/LAYERS generally lets the network…

Accepted Answer

Learn more complex, non-linear decision boundaries — Greater capacity can carve more intricate boundaries (e.g. for spiral data), but risks overfitting if excessive.

Question 3

The 'spiral' dataset is the hardest because it requires…

Accepted Answer

A highly non-linear boundary, needing enough depth/neurons and good features — Spirals can't be separated linearly; the playground shows you need sufficient hidden units (and often engineered features) to solve it.

Question 4

Choosing input FEATURES like x₁², x₂², or sin(x) demonstrates…

Accepted Answer

How feature engineering can make a hard problem easier for the network — Adding non-linear features can let even a small network separate complex data, illustrating the value of good representations.

Question 5

Setting the LEARNING RATE too high in the playground causes…

Accepted Answer

The loss to oscillate or diverge instead of smoothly decreasing — You can visually see large learning rates make the boundary thrash; too-small rates converge very slowly.

Question 6

The thickness/color of the connections between neurons represents…

Accepted Answer

The magnitude and sign of the learned weights on each connection — The visualization encodes weight values, letting you watch them change as the network learns.

Question 7

If TRAINING loss keeps dropping but TEST loss rises, the network is…

Accepted Answer

Overfitting to training noise — The playground shows separate train/test loss; a growing gap signals overfitting, which regularization can reduce.

Question 8

Enabling L1/L2 REGULARIZATION in the playground tends to…

Accepted Answer

Shrink large weight magnitudes via a penalty term, producing smoother decision boundaries — Regularization penalizes large weights; you can watch the decision boundary become smoother and connections fade.

Question 9

Switching the ACTIVATION from ReLU to a linear activation would make a deep network…

Accepted Answer

Only able to learn linear decision boundaries, regardless of its depth — Without non-linear activations, stacked layers collapse to one linear map — the playground makes this limitation visible.

Question 10

The colored background regions in the output represent the network's…

Accepted Answer

Predicted class or value for every location across the entire input space — Each region's color is the network's prediction there; watching it reshape during training shows learning in action.

What does this dataset need?

Test your knowledge

TensorFlow Playground lets you train a small neural network…

Interview Questions

References & Further Reading

Love this visualization?