A1065
Title: Semiparametric analysis of deep ordinal choice models
Authors: Yiwei Fan - Beijing Institute of Technology (China) [presenting]
Abstract: Deep learning has achieved considerable success across various application domains, coupled with notable advancements in its theoretical foundations. Despite these strides, the exploration of deep learning in the context of the maximum rank correlation (MRC) estimator remains relatively limited. A smoothed MRC estimator is introduced for the ordinal choice model, which integrates a linear function for interpretation and a nonlinear function fitted using deep neural networks. A two-step algorithm is designed for estimation, which keeps the order relation among outputs without the parallelism assumption. Under regular conditions, statistical properties are established for the smoothed MRC estimator, including identification, convergence rate, and minimax-optimality, where the number of categories is allowed to increase with the sample size. Simulations and extensive real-world applications demonstrate the advantages of the proposed method in terms of classification accuracy and interpretability.
