- This event has passed.
Introduction to Deep Neural Network using R
June 28, 2020 @ 9:00 am - 10:00 am
Title: “Introduction to Deep Neural Network using R”
Speaker: Tuhin Sheikh, ISRT, University of Dhaka
and PhD candidate at the University of Connecticut, USA,
Date and Time: Sunday, June 28, 2020 at 9.00AM
—————————–Summary————————————————
The deep neural network (DNN) modelling has been considered to be a thriving topic in recent years. The DNN can be considered as a generalization of traditional regression analysis. Considering a particular objective of predicting output, traditional regression analysis extracts low level features based on the observed input covariates. However, when we deal with high dimensional data and a numerous input features, low level feature extraction often leads to low prediction accuracy. The DNN on the other hand, has been found to effective in extracting high level feature with promising prediction accuracy. In DNN, we assume that similar to the neural system, the input covariates go through different neurons at different layers until it reaches the final output layer. The higher the number of middle layers, the deeper the network is. If there is no (hidden) layers in the middle, it generalizes to only input and output layer as in traditional regression. Like regression, DNN requires a loss function and criterion for minimization. The key difference would be, extraction of hidden features and connecting those to the final output prediction.
Due to the advancement of computer algorithms and emergence of interesting data, this research field has been found to interesting in present times. Many big companies (e.g. Google, Facebook, Boehringer Ingelheim, etc.) have been practicing deep neural network modelling due to the satisfactory performance. In the past, mostly computer scientists and engineers contributed in this field. However, the attractive mathematical foundation behind this interesting methodology got the attention of the Statisticians recently. As Statisticians, there are huge scopes to contribute to this emerging field through statistical innovation. In this session, the audience can expect the discussion on background and basic mathematical foundation of DNN. Also, I will introduce an R package “Keras”, which can be used to work with deep neural network. At the end of the discussion on this interesting topic, I will spend some time discussing higher study experience in USA.
