Nonparametric Estimation of Mixing Multinomial Logit Models by Kernel Machine

ABSTRACT

Non-parametric mixing multinomial logits are crucial in many classification problems given data are generated by mixture of unknown multinomial logits. We first show that any estimation methods require exponential amount of data, which makes the problem intractable. However, by a small modification of the original problem, we can have a non-parametric method based on tensors decomposition of kernel machines.  We can also show that our model can achieves minimax convergence rate to the true mixture components as the data volume increases.

SPEAKER BIO

Liang Ding is an Assistant Professor in the School of Data Science at Fudan University. Prior to Fudan, he was a postdoctoral research associate joint appointed by the ISEN department and TAMIDS at Texas A&M University. He got his Ph.D. in operations research from the Hong Kong University of Science and Technology. His research interests lie in simulation optimization, kernel methods, and statistical learning.