Fibonacci Super Learners and stacked generalization

Hybridize geometry-driven Fibonacci weights with stacked generalization and Super Learner by constraining the meta-learner to the Fibonacci conic hull or using Fibonacci weights as a structured prior over stacking coefficients, and analyze the resulting asymptotic oracle properties under golden-ratio regularization.

Background

Stacking and Super Learner are cited as complementary aggregation frameworks that optimize convex combinations using cross-validated risk. The paper’s geometry-driven Fibonacci weighting offers a deterministic alternative rooted in golden-ratio structure.

The Future Work section suggests hybridizing these approaches, such as restricting the meta-learner to the Fibonacci conic hull or imposing Fibonacci weights as a prior over stacking coefficients to blend oracle guarantees with structured weighting.

References

Several directions remain open and are, in our view, both challenging and promising.

— On Fibonacci Ensembles: An Alternative Approach to Ensemble Learning Inspired by the Timeless Architecture of the Golden Ratio  (2512.22284 - Fokoué, 25 Dec 2025) in Section “Future Work: From One-Dimensional Harmony to High Dimensional Practice”