Saturday 2:30 PM–3:00 PM in C01

Machine Learning Architectures

Shagun Sodhani

Audience level:
Intermediate

Description

We would talk about and implement some common machine learning architectures and building blocks which can be applied to a variety of use cases. The topics include Siamese networks, Triplet Networks, Skip connections, Batch Normalization and Dropout. We would use the Duplicate Question Dataset from Quora to demo these architectures.

Abstract

We would talk about and implement some common machine learning architectures and building blocks which can be applied to a variety of use cases. I would show all the techniques as part of steps we take for solving the Duplicate Question Detection problem (based on Quora Dataset ) so that the audience understands how these ideas complement each other.

The topics to be discussed include:

Siamese networks - Useful for finding similarity between two comparable objects (text, image etc)

Triplet Networks - Extension of Siamese Networks

Skip/Residual connections - Allow for training deeper networks

Batch Normalization - Accelerate the training of deep neural networks

Dropout - An amazingly simple regularisation trick

Prerequisites:

Attendees should have a basic idea of machine learning and be comfortable with Python

Content URLs:

The code and presentation (along with detailed explanation) would be available here.

Subscribe to Receive PyData Updates

Subscribe

Tickets

Get Now