Deep Learning, and Deep Networks for Natural Language Processing


Details
The Talk
Theory and justification behind deep learning
- Biological justification
- Learning distributed representations
- Transfer learning
Unsupervised pre-training
Benchmark results
Current Approaches - Restricted Boltzman Machines and Deep Belief Networks
- Auto-Encoders (contractive, sparse, and denoising)
- Recursive Auto-Encoders used for NLP classification (based on the Stanford group's work led by Richard Socher).
Future Topics
About Simon Hughes
Having spent the last 8 years working as a software developer, mainly in the finance industry, Simon recently started a new position as a Data Scientist working for Dice.com, the recruiting website. They are currently investigating ways to apply machine learning and NLP techniques to provide insight into the recruiting and hiring process. He is also working part-time on a PhD at DePaul university where he and his colleagues have had 3 papers published on the topic of automating the coding of student essays, detecting argumentation and claims within essay sentences. The end goal of this research is to extract the information from essays that would be necessary to create an automated essay tutor to teach students to write better essays. His current focus is on applying deep learning algorithms to improve the classification accuracy on their dataset over more traditional approaches.

Deep Learning, and Deep Networks for Natural Language Processing