Deep Learning with Fast.Ai

Deep learning is a subfield of machine learning which is stirred by the structure and function of the brain called artificial neural networks. Deep learning have been applied to many fields, computer vision, natural language processing, bioinformatics, etc. Deep learning is one of the highly demanding skills in tech, and mastering deep learning open numerous new career opportunities. Fast.ai is a high level programming framework for deep learning developed by Jeremy Howard. It provides a simplified programming interface wrapper over Pytorch. In this one day course you will learn how to use Fast.ai for deep learning to solve different practical problems. The topics includes:

• Deep Neural Network (DNN) and Convolutional Neural Network (CNN)
• Image Recognition and Classification with Fast.ai
• Tuning hyper-parameters to improve DNN and CNN
• Fast.ai for structured/tabular data, e.g. time-series sales data
• Recurrent Neural Network (RNN)

Module 1 Introduction to Fast.ai

• What is Fast.ai
• Install Fast.ai
• Fast.ai vs Pytorch

Module 2 Deep Learning

• What is Deep Learning?
• Families of Deep Learning
• Simple Neuron, Perceptron, and Multilayer Perceptron
• Learning Rule
• Autoencoders
• Regression vs Classification

Module 3 Convolutional Neural Network (CNN)

• What is CNN?
• Difference between Fully connected NN and CNN.
• Stride, Padding, and Pooling
• ReLU Transfer Function

Module 4 Deep CNN and its application to Image Classification

• Vgg16, ResNet34/50, GoogleNet, and DenseNet121
• Transfer Learning
• Two class image classification (Dog vs Cat data) with Fast.ai
• Multi-class image classification (Dog bread data) with Fast.ai
• Multi-labels image classification (Satellite image data) with Fast.ai

Module 5 Improving Deep Neural Network

• Batch Normalization, Dropout
• Tuning Hyper-parameters to improve performance of NN.
• Learning Rate Annealing
• Different Cost Functions

Module 6 Deep learning for structured/tabular data with Fast.ai

• Structured/tabular data
• Deep Learning for time-series data
• Language modelling with Fast.ai

Module 7 Recurrent Neural Network (RNN)

• What is RNN?
• Architecture of RNN
• Long Short Term Memory (LSTM)
• Movie Review data and RNN

Sanjay is a Director at Alpha Alternatives Advisors, a Fund Management Company in Singapore, responsible for execution of investment mandates. Sanjay has over 15 years’ experience in Investment Banking and Asset Management across New York, London and Singapore.

Sanjay started his career in Finance as a trading desk quant at Lehman Brothers, New York. He worked closely with traders to develop and test equity derivatives trading strategies to exploit arbitrage, mean-reversion and other fundamental relationships. He was instrumental in building a Risk Management Advisory practice (Quant Advisory) with Ernst & Young in New York and London. He executed several key credit & market risk mandates with major investment banks like Morgan Stanley and Citigroup Subsequently he changed roles to become a Capital Markets Banker with Standard Chartered Bank (SCB) from 2007 to 2013 initially based in London before moving to Singapore in 2010.

Sanjay holds a Masters in Computational Finance (Financial Engineering) from Tepper School of Business, Carnegie Mellon University, Pittsburgh USA. He also holds a Bachelors in Electrical Engineering (Hons).

Sanjay is an engineer at heart and has extensive programming experience beginning with COBOL on mainframes, C/C++ on Unix and subsequently on modern languages like Java. He was one of the first Sun Certified Java Developer in 2000. His quantitative specialities include:

• Machine Learning/Deep-learning Toolkits: PyTorch and Tensorflow, Python and its associated popular libraries like Pandas, Scikit-learn, Excel/VBA
• Visualization and Statistics using Python, R, Tableau and Excel/VBA
• Algorithms for machine learning: Regression Modeling, Decision Trees, Clustering, Text Mining, Simulation methods and Time Series