Skip to main content

Loading Events

« All Events

  • This event has passed.

Numerical Analysis Seminar: Elizabeth Newman, Emory University, How to Train Better: Exploiting the Separability of Deep Neural Networks

April 18 | 3:00 pm - 4:00 pm EDT

Deep neural networks (DNNs) have gained undeniable success as high-dimensional function approximators in countless applications. However, there is a significant hidden cost behind triumphs – the cost of training. Typically, DNN training is posed as a stochastic optimization problem with respect to the learnable DNN weights. With millions of weights, a non-convex and non-smooth objective function, and many hyperparameters to tune, solving the training problem well is no easy task.

In this talk, we will make DNN training easier by exploiting the separability of common DNN architectures; that is, the weights of the final layer of the DNN are applied linearly. We will leverage this linearity in two ways. First, we will approximate the stochastic optimization problem deterministically via a sample average approximation. In this setting, we can eliminate the linear weights through variable projection (i.e., partial optimization). Second, in the stochastic optimization setting, we will consider a powerful iterative sampling approach to update the linear weights, which notably incorporates automatic regularization parameter selection methods. Throughout the talk, we will demonstrate the efficacy of these two approaches through numerical examples.


April 18
3:00 pm - 4:00 pm EDT
Event Category:


SAS 4201