Our approach, titled Scalable Polynomial Additive Models (SP AM) is effortlessly scalable and models all higher-order feature interactions without a combinatorial parameter explosion. SP AM outperforms all current interpretable approaches, and matches DNN/XGBoost performance on a series of real-world benchmarks with up to hundreds of thousands of features.

Sequential data in our everyday life is often hierarchically structured.

This paper considers online convex optimization with hard constraints and analyzes achievable regret and cumulative hard constraint violation (violation for short).