Enabling Powerful NLP Pipelines with Transfer Learning

Abstract: Traditionally, NLP problems are complex tasks that requires large manually labelled datasets and complex ML models. Transfer learning and Deep Learning have made a breakthrough in Computer Vision, and a similar class of models are set to revolutionise NLP. Lars Hulstaert will discuss how models like ElMo, GPT-2 and UlmFit will reduce the entry barrier for extremely powerful NLP pipelines.
Topics include:
A definition of transfer learning and language models.
New trends in language models: statistical models vs. deep learning.
A comparison of the state-of-the-art in language models.
A set of guidelines to enable transfer learning for your NLP problems.
A look ahead.
The takeaway for the audience:
- Understand how to build strong NLP pipelines
- Learn how transfer learning can be applied to NLP
- Enable teams to quickly validate NLP problems.

Bio: Lars Hulstaert is a Data Scientist at Microsoft, where he helps delivering machine learning projects at enterprise customers. Lars serves as an expect resource on machine learning on unstructured data and deep learning. Before joining Microsoft, Lars studied Computer Science Engineering at Ghent University and Machine Learning at Cambridge University. His current focus is on computer vision and robotics although he is really passionate about transfer learning techniques that enable others to build robust pipelines with little data.