EcoSta 2022: Start Registration
View Submission - EcoSta2022
A0821
Title: Understanding deep learning via statistical modelling approaches Authors:  Il Do Ha - Pukyong National University (Korea, South) [presenting]
Jihun Kim - Pukyong National University (Korea, South)
Abstract: Recently, deep learning (DL) have provided breakthrough results for prediction problems including classification for a wide variety of applications. In particular, the core architectures that currently dominate the DL are deep feed-forward neural networks (DFNN), CNN, RNN, LSTM, AE and GAN, etc. The DL models are represented as structured neural networks consisting of three layers (input, hidden and output layers) for constructing (or modelling) the functional relationship between input and output variables, and the main goal is to find a nonlinear predictor of the output $Y$ given the input $X$. The output models of DL can be expressed as structured mean models, leading that the estimation of such mean provides the prediction of $Y$. It is thus interested to study the DL in terms of statistical perspective. The DL models can be viewed as a highly nonlinear and semi-parametric generalization of statistical models such as the generalized linear model (GLM). The fitting (i.e. learning) of DL models based on train data is usually implemented using likelihood-based methods including the construction of loss function or regularization. We present how to understand the DL models via the GLM framework, and then extend it to survival models allowing for censoring and to random-effect models, with practical examples.