Jacobian Rank Collapse in Decision-Focused Learning

Battle Count: 50

Highly relevant as it directly addresses the efficacy of Decision-Focused Learning in portfolio optimization and index tracking, providing critical insights into when task-aware training outperforms traditional MSE-based prediction.

Requires implementing differentiable optimization layers (e.g., via cvxpylayers), managing complex training protocols with validation tuning, and conducting extensive controlled experiments across multiple domains.

The paper provides a detailed reproducibility statement, including saved results, editable figures, numerical checks, and training source snapshots. It distinguishes between corrected equity evaluations, archived diagnostics, and new synthetic runs. However, historical real-data training has not been independently repeated in this release.