DOS Seminar - Joseph Paat and Amitabh Basu

Primary tabs

Joseph Paat (https://sites.google.com/site/josephspaat/)

Title: How to choose what you lift

Abstract: One way to generate valid cuts (that is, valid inequalities) for a mixed-integer set is to first relax any integrality constraints and create a cut for the continuous relaxation. However, as this approach ignores integrality information, cuts generated in this way may be weak. To remedy this, one may hope to somehow reintroduce integrality information in an attempt to strengthen the cut. This process of reimposing integrality on a variable is referred to as lifting the variable. In this talk we explore the idea of lifting from the viewpoint of cut-generating functions. We examine questions such as "Does lifting one variable produce a stronger cut than lifting another?" and "How much strength is gained from lifting a single variable?" This work was done in collaboration with Santanu Dey and Amitabh Basu.



Amitabh Basu (http://www.ams.jhu.edu/~abasu9/)

Title: Understanding Deep Neural Networks with Rectified Linear Units

Abstract: In this paper we investigate the family of functions representable by deep neural networks (DNN) with rectified linear units (ReLU). We give the first-ever polynomial time (in the size of data) algorithm to train a ReLU DNN with one hidden layer to global optimality. This follows from our complete characterization of the ReLU DNN function class whereby we show that a  function is representable by a ReLU DNN if and only if it is a continuous piecewise linear function. The main tool used to prove this characterization is an elegant result from tropical geometry. Further, for the  case, we show that a single hidden layer suffices to express all piecewise linear functions, and we give tight bounds for the size of such a ReLU DNN. We follow up with gap results showing that there is a smoothly parameterized family of  "hard" functions that lead to an exponential blow-up in size, if the number of layers is decreased by a small amount. An example consequence of our gap theorem is that for every natural number , there exists a function representable by a ReLU DNN with depth  and total size , such that any ReLU DNN with depth at most  will require at least  total nodes. Finally, we construct a family of  functions for  (also smoothly parameterized), whose number of affine pieces scales exponentially with the dimension  at any fixed size and depth. To the best of our knowledge, such a construction with exponential dependence on  has not been achieved by previous families of "hard" functions in the neural nets literature. This construction utilizes the theory of zonotopes from polyhedral theory.


  • Workflow Status: Published
  • Created By: Anita Race
  • Created: 02/16/2017
  • Modified By: Fletcher Moore
  • Modified: 04/13/2017


No keywords were submitted.

Target Audience

No target audience selected.