Events login

Main Events Calendar

March 25, 2019 | 2:30 p.m. - 3:30 p.m.
Category: Lecture
Location: Engineering, College of #1507 | Map
5050 Anthony Wayne
Detroit, MI 48202
Cost: Free

Speaker: Gilbert Strang (MIT)

Title: The Functions of Deep Learning


Deep learning refers to the creation of a function F(v) that matches the known outputs on a training set of data, for example on images v of digits or letters. The key point (and the magic) is that F should also succeed on unseen data v from the same population. The function is normally continuous and piecewise linear:  a composition of simple functions of the form  max ((Av + b) , 0). The success of modern machine learning is to choose matrices A and vectors b that minimize the final error ("loss function")between F(v) and the known output from v --- this large optimization is reduced for Convolutional Nets when A uses the same weights at all points of the image :  a banded Toeplitz matrix at each layer of F.

For more information about this event, please contact Department of Mathematics at 3135772479 or