ML Math Lab — Gradient Descent
Guide the energy ball to the valley bottom (minimum loss) — without overshooting.
Mode: Explore
Goal: Converge smoothly
Score: 0
Section 1 — What you will learn
By the end you will understand
  • Learning Rate (α): step size — how big the ball’s “move” is each step.
  • Gradient (slope): tells which direction is downhill and how steep.
  • Convergence: when the ball reaches the bottom and stops.
  • Overshooting: when α is too big, ball jumps past the bottom.
  • Momentum (β): gives “inertia” — ball keeps moving even when slope is small.
  • Damping: friction — removes extra speed so it can settle.
Section 2 — Controls
Different valleys teach: “one α does NOT fit all.”
Low α = tiny steps. High α = big jumps.
Momentum makes the ball keep rolling even when slope becomes small.
If ball keeps bouncing, increase damping a bit.
Try starting on different sides to see symmetry and traps.
Kids Mission: Make the ball reach the bottom in ≤ 25 steps without exploding out of the valley.

Section 3 — Live stats
Current Loss (Altitude)
Slope (Gradient)
Step Count
Direction
Step Size (α · |grad|)
Speed (Momentum)
Status
Explore: move α / β / damping / start point.
Mini Mission
Make it converge in under 25 steps.

Section 4 — Math (The Engine)
How the ball updates θ Watch “ghost dot” next step
Loading…
🎮 Kid Coach
Loading…
Tip: If it bounces forever, reduce α or increase damping.
Live explanation
Click START TRAINING to begin. Use α + Momentum to control the ball.
Game HUD
Explore: learn what α, β, damping do to the ball.
Overshoot Meter
If loss increases often → α too high.
Momentum Meter
High speed near bottom = bouncing risk.
Convergence Meter
When slope ~ 0 and speed ~ 0, you win.
Quick actions
Pro tip: Ghost dot shows the next landing spot. If it jumps across the bottom → overshoot.