SC Colloquium: "Representation Learning - The Hidden Universe of Data"
We live in a world where data is generated at unprecedented rates. According to IBM, we create 2.5 exabytes of data daily. This data is a treasure trove of knowledge and information - however the universe it lives in is dauntingly complex and our current algorithms are only beginning to untangle its mysteries. Much of our success in analyzing and understanding this data relies heavily on its representation. For example, most would find little trouble in dividing two large numbers in decimal form. However, given the numbers as roman numerals, division may prove more taxing. In this talk, we will present a cursory overview of the field of representation learning. We will examine its origins, its capabilities, and its future. We will conclude with some of the research questions under examination in the Department of Scientific Computing's Intelligence Lab.