MAD
Minimum Ascent Descent — multi-minimum optimizer
Four-phase C++ optimizer that exhaustively finds all global minima of a differentiable 2D loss function. Standard gradient descent collapses to one minimum; MAD uses minimum-ascent trajectories, pass-point stacking, and directional exclusion sets to systematically escape and catalogue every basin. Validated on Himmelblau, Rastrigin, Ackley, and Beale. Collaborations welcome — reach out if this direction interests you.