A dance presentation of how machines learn to "think" like humans by Apoorv Agarwal in collaboration with choreographer Caitlin Trainor.
In this dance presentation we are explaining the idea of how machines learn from data. For example, how would a machine learn to answer a Jeopardy! question like "4 letter word for a vantage point or a belief"? We are presenting a relatively new machine learning paradigm, which is deeply rooted in a strong mathematical foundation. It is called Kernel Learning. More specifically, we are presenting how a perceptron in its dual form uses convolution kernels to learn to differentiate between two categories of objects.
In my thesis, I am using this machine learning paradigm for two applications: 1) automatic social network extraction from fiction and non-fiction text and 2) sentiment analysis of social media data like tweets. Following are brief explanations of these two applications.
Friendships and other social relations are conveyed through meta-data such as self-declared friendship links and emails. But a more accurate social network is present in the content of conversations (or emails). People express: who they meet, talk to, think about etc. using language. The goal of my thesis is to automatically detect such interaction links present in text. For example, given a text like Alice in Wonderland, we are able to extract a network of characters (Alice, Rabbit, Queen etc.) and how they interact. From network analysis of this network, it becomes clear, for instance, that even though the Rabbit is mentioned many times in the text, it is not an influential character.
Microblogging websites have evolved to become a source of varied kind of information. This is due to nature of microblogs on which people post real time messages about their opinions on a variety of topics, discuss current issues, complain, and express positive sentiment for products they use in daily life. In fact, companies manufacturing such products have started to poll these microblogs to get a sense of general sentiment for their product. Many times these companies study user reactions and reply to users on microblogs. I am using the machine learning paradigm we present in this dance presentation to build technology to detect and summarize an overall sentiment.
This project is done in collaboration with choreographer Caitlin Trainor.
The film is shot by Andrew Gladstone and edited by Heather Kemp.
Cover photograph is by Lucas Chilczuk.