Home▸Data Science▸Gradient Surgery: Projecting Away Conflicting Task Gradients (2D)

Gradient Surgery: Projecting Away Conflicting Task Gradients (2D)

Parameter-space trajectories of plain summed gradients and projected (PCGrad) gradients on two quadratic task losses, with both loss curves beneath. Same shared model and checks as the 3D page.

Data Science2DModerate60 FPS📱 Mobile-adapted🟠 Blender⇄ 3D version
Gradient Surgery: Projecting Away Conflicting Task Gradients (2D) ↗ Open standalone

Two tasks pull one shared parameter vector in opposite directions. Project each gradient onto the normal plane of the other whenever they conflict, and compare the path with plain gradient summing. Change the curvature mismatch, the distance between the task optima and the step size, then read the conflict fraction, both losses and the Pareto stationarity gap. A Blender-built plotting easel with exact orthogonality, descent and finite-difference checks built in.

⚙ Under the hood

Parameter-space trajectories of plain summed gradients and projected (PCGrad) gradients on two quadratic task losses, with both loss curves beneath. Same shared model and checks as the 3D page.

Gradient SurgeryPCGradMulti-Task LearningConflicting GradientsPareto Stationarity

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install

What did you find?

Add reproduction steps (optional)