Projection Set Algorithms

• Proximal gradient methods are a powerful technique to solve non-convex inverse problems
• Problem: find the intersection of two sets
  • Catch: only allowed to operate on one set a time

Two Convex Sets

• Two convex sets → "alternating projections" guaranteed to converge

Two Non-convex Sets

• Two non-convex sets → "generalized projections" show more robust convergence

Multiple Non-convex Sets

• Multiple non-convex sets → require consistency across parallel projections