Timezone: »

What training reveals about neural network complexity
Andreas Loukas · Marinos Poiitis · Stefanie Jegelka

Thu Dec 09 04:30 PM -- 06:00 PM (PST) @

This work explores the Benevolent Training Hypothesis (BTH) which argues that the complexity of the function a deep neural network (NN) is learning can be deduced by its training dynamics. Our analysis provides evidence for BTH by relating the NN's Lipschitz constant at different regions of the input space with the behavior of the stochastic training procedure. We first observe that the Lipschitz constant close to the training data affects various aspects of the parameter trajectory, with more complex networks having a longer trajectory, bigger variance, and often veering further from their initialization. We then show that NNs whose 1st layer bias is trained more steadily (i.e., slowly and with little variation) have bounded complexity even in regions of the input space that are far from any training point. Finally, we find that steady training with Dropout implies a training- and data-dependent generalization bound that grows poly-logarithmically with the number of parameters. Overall, our results support the intuition that good training behavior can be a useful bias towards good generalization.

Author Information

Andreas Loukas (EPFL, MIT)

Researcher fascinated by graphs and machine learning.

Marinos Poiitis (Aristotle University of Thessaloniki)
Stefanie Jegelka (MIT)

Stefanie Jegelka is an X-Consortium Career Development Assistant Professor in the Department of EECS at MIT. She is a member of the Computer Science and AI Lab (CSAIL), the Center for Statistics and an affiliate of the Institute for Data, Systems and Society and the Operations Research Center. Before joining MIT, she was a postdoctoral researcher at UC Berkeley, and obtained her PhD from ETH Zurich and the Max Planck Institute for Intelligent Systems. Stefanie has received a Sloan Research Fellowship, an NSF CAREER Award, a DARPA Young Faculty Award, the German Pattern Recognition Award and a Best Paper Award at the International Conference for Machine Learning (ICML). Her research interests span the theory and practice of algorithmic machine learning.

More from the Same Authors