Microsoft Reactor

What is this session about?
Language Modelling is the task of predicting what word comes next. In formal definition, given a sequence of words, a language model computes the probability of the next word. Language models are best used for part of speech tagging, name entity recognition, sentence classification, text summarisation, text generation, and question answering. In this session we briefly review n-gram language models and Recurrent Neural Network (RNN) which are the traditional approaches to building language models and have the vanishing gradients problem. We then talk about the Transformers architects that have significantly evolved from previous ML architectures. In the light of transformers-based architects, we introduce the Generative Pre-trained Transformers (GPT) models by OpenAI and will review their advantages and disadvantages. We finish the session by examples of accessing the API of Microsoft Turing Natural Language Generation (T-NLG) with opportunity of hands-on practices.

Who is it aimed at?
The session is aimed at new data scientists with some introductory level of NLP and established data scientists interested in developing simple models using the state of the arts language models APIs.

Why should I attend?
If you wish to learn more on the following technologies or techniques this is the session for you:
Introduction to RNNs architects
Review of Transformers_based architects
Overview of GPT-1, GPT-2, GPT-3
Hands-on-code with T_NLG API

For more on this series, visit https://developer.microsoft.com/en-us/reactor/eventseries/NaturalLanguage

Speaker info:
Lida Ghahremanlou - Cloud Solution Architect - Microsoft

Dr. Lida Ghahremanlou is an AI specialist and a Cloud Solution Architect at Microsoft, Data & AI Azure team from Customer Success Unit (CSU), working closely with customers and partners on providing AI and Machine learning solutions using Azure cloud technologies. Prior to joining Microsoft, she was a research scientist at IBM-Research in Australia where her focus was on Cognitive Computing and Semantic Technologies. Lida is passionate in applying her skills and experience within NLP applications and solving real world problems of language understanding.