Constrained lasso

\(\min_{\|\mathbf x\|_1\leq R_1}\tfrac12\|A\mathbf x-\mathbf b\|_2^2\)

Iteration t
Loss f(xₜ)
Coefficients xₜ
‖xₜ‖₁ / R₁

Same quadratic curvature and orientation; unconstrained minimum shifted to (4,1) so the constraint matters. The L1 ball is the diamond; gray is infeasible. Change R₁ to see the effect on the coefficients. This is constrained lasso; the projection threshold is determined by the radius.