🎯
Physics Playground
From: Observations, Not Numbers•View in lesson →
💡
Data doesn't exist in a vacuum—it represents something real. In this playground, you can experiment with physics simulations and see how measurements translate into the data that machines learn from. Observe cause and effect, and understand why data quality matters.
0
Steps Taken
0.000
Current Loss
100
Position (Parameter)
How big each step is
Small stepsLarge steps
How much the ball "remembers" previous direction
No memoryHigh memory
Try These Scenarios:
What's Happening?
- '
- ⛰️ The landscape is the “loss function" - height = how wrong the model is
- ⬇️ Gradient descent = rolling downhill to find lowest loss
- 🔴 The red arrow shows the gradient (which direction is downhill)
- 🟡 The yellow path shows all the steps taken during training
Learning Rate: Too small = slow progress. Too large = ball overshoots and bounces around.
Momentum: Helps ball roll through small bumps and reach deeper valleys.
Experiment as much as you like; there is no progress to lose.