Outline of SINGA Tutorial for SDSC

sdsc.md

Deep Learning Practice with Apache SINGA

Instructor

WANG Wei, Assistant Professor, School of Computing https://www.comp.nus.edu.sg/~wangwei/

Overview

In this short-course, you will learn deep learning through practices with the Apache SINGA library. First, deep learning basics, including neural network layers and training algorithms, will be introduced by training a simple model from scratch. Next, you will learn to do transfer learning based on pretrained deep learning models for practical computer vision and natural language processing tasks. Some tricsk for speeding up training and improving accuracy will also be covered. After this course, you will be equiped with the skills and knowledge for applying deep leanring for practical applications.

Outline

Session I

1. Quick introduction to deep learning and existing libraries

2. Overview of Apache SINGA and installation

3. Image classification from scratch

   3.1 Demo

   3.2 Preparing data

   3.3 Constructing a convolutional neural network (CNN)

   3.4 Training the model

   3.5 Predicting for new samples

Session II

4. Transfer learning for food image classification

   4.1 Demo

   4.2 Preparing data

   4.3 Loading a pre-trained CNN model

   4.4 Updating the model structure

   4.5 Advanced training algorithms and distributed training

   4.6 Deploying the model for online serving with Flask

Session III

5. Transfer learning for question answering

   5.1 Demo

   5.2 Preparing data

   5.3 Loading a pre-trained question answering model (LSTM-based)

   5.4 Updating the model structure

   5.5 Training the model

   5.6 Deploying the model for online serving (on Telegram)

Prerequisites

1. With a Google account to access Google Colab notebooks https://colab.research.google.com/
2. Faimilar with Python and Numpy. Here is a quick review: https://colab.research.google.com/github/cs231n/cs231n.github.io/blob/master/python-colab.ipynb (Just refer to the Python and Numpy sections).