What Can I Do With "Deep Learning"?

01:15 PM - 01:40 PM on August 15, 2015, Room 701

Kyle Kastner

Audience level:

intermediate

Watch:

http://youtu.be/POCQBk1oBzI

Description

"Deep learning" is a recent rebranding and mixing of old and new methods in neural networks, graphical modeling, and optimization. We will discuss the applications of these approaches, how these methods are different than others for machine learning, and what recent advances in the field mean for people trying to solve problems in the real world.

Abstract

"Deep learning" is a recent rebranding and mixing of old and new methods in neural networks, graphical modeling, and optimization. Recent advances in feedforward neural networks, convolutional networks, and recurrent networks coupled with vast improvements in parallel computing have enabled the ability to process huge amounts of data. Deep learning techniques allow the flexibility to learn from data with little or no feature engineering, and the set of methods which comprise "deep learning" can be used with little modification across tasks in different domains. The ability to learn complex functions from raw data has made an outsized impact on applied machine learning, and applications of these methods have revolutionized speech, text and image processing.

We will discuss the applications of these approaches, how these methods are different than others for machine learning, and what recent advances in the field mean for people trying to solve problems in the real world.