Lab · c17.s01.L1
Gradient descent 2D
The negative gradient is the steepest downhill direction at the current point, but on a stretched bowl that local arrow leads across the contours, not straight to the bottom.
Question: How does a gradient point to a better answer?
Frame 1 of 6: StartStart at (2.0, 1.5) with loss 6.5000. The gradient is (2.00, 6.00), so the russet arrow points across the grey contours at a right angle: the steepest downhill direction at this point, and not at the centre of the bowl.
Start. Start at (2.0, 1.5) with loss 6.5000. The gradient is (2.00, 6.00), so the russet arrow points across the grey contours at a right angle: the steepest downhill direction at this point, and not at the centre of the bowl.
Keyboard: Space plays or pauses, Left and Right step, Home goes to the first frame, End goes to the last frame.
Text description of this lab
A stretched bowl from above: five nested grey ellipses around the lowest point, labelled 0.5, 1, 2, 4 and 6.5. A navy path steps downhill from the upper right, lowering the loss at every step. A russet arrow shows the negative gradient there, crossing the contours at a right angle. The path zigzags across the long axis because the arrow is steepest only where it stands. A side panel prints the current point, loss and gradient.
Return to the section that hosts this lab
Caption status: re-simulated. Every lab declares one of published, re-simulated, schematic, synthetic or interpolated.