SOFTWARE 2.0
Using Multitask AI and Natural Language Processing to Transform your Organization

by Russell Jurney download a PDF brochure

Description

AI software is eating the world. Disruption has begun to affect every sector and industry. Your organization is either ahead of the curve or will fall rapidly behind as front runners pull away. To achieve this, you don’t just need data, you need data labeled for the outcomes you want to predict to drive your organization’s most critical operations. However, much of this data is locked away inside of documents as text and is not labeled for the required task, which can require a large investment of time and money.

This course teaches a technique called weak supervision that uses data programming to generate weak labels that are combined by an unsupervised generative model to produce strong labels for your data. Automatically. These labels are then used to train a multitask neural network to solve the core, often related problems of the business. This multitask model replaces the software repository as the core technology on which the company runs. Software engineers, machine learning engineers and data scientists then contribute alterations to this model to improve performance or to solve an additional task. This is Software 2.0, and it is how Google, Apple, Facebook and Microsoft operate.

This class will teach the fundamentals of Natural Language Processing (NLP) for Software 2.0: core NLP skills, neural networks, multitask learning and weak supervision. We will show how to process text to extract structured data, how to use Tensorflow 2.0 to build neural network models, how to use Snorkel and data programming to perform weak supervision and cheaply label data, how to combine multiple models into a single multitask model and how to make use of existing state of the art networks to solve problems.

What you will learn

• How the Software 2.0 model operates to transform a business using AI
• Natural Language Processing using Python and spaCy
• How to build and train neural networks for natural language processing using Tensorflow 2.0
• How to use Snorkel to programmatically label data using Weak Supervision
• How to use Distant Supervision to label data using internal and external knowledge bases
• How to use Transfer Learning to rapidly build new models
• How to employ state of the art network architectures in Tensorflow Hub
• How to build multi-task models to solve more than one related problems
• How to perform Named Entity Resolution (NER) to extract structured data from text
• How to classify text documents

Main Topics

• Weakly Supervised Learning
• Collecting and Processing Data
• Transfer Learning
• Weak Supervision
• Semi-Supervised Learning
• Distant Supervision
• Multi-Task Learning