Assess the learner's ability to create an original accelerated proximal-gradient algorithm for large-scale nonconvex, nonsmooth optimization problems with affine constraints. Require clear statement of model and assumptions (e.g., Lipschitz gradients, prox-operators, constraint qualification), full pseudocode, discussion of implementation and scalability for large-scale data, rigorous convergence analysis (stationary-point guarantees, subsequence vs. whole-sequence convergence), and explicit complexity bounds (iteration and oracle complexity). Expect comparison to baseline methods, numerical experiment design, and explanation of how affine constraints are handled (e.g., projections, augmented Lagrangian, dual updates).