Mittagsseminar (in cooperation with J. Lengler, A. Steger, and D. Steurer)

Date and Time: Tuesday, May 26, 2020, 12:15 pm

Duration: 30 minutes

Location: Zoom: conference room

Speaker: Tim Taubner

VC dimension of neural networks with rectified activations

The talk was recorded. Use the following link to access it: https://ethz.zoom.us/rec/share/xfVrMoHM6URIZp3k4UrWW6QxH467X6a8hHcWr_ZYnU-Jm9bTrJcx5f0IOq1xZG90

The Vapnik-Chervonenkis (VC) dimension is one way to measure the capacity of a family of classifiers. (Deep) neural network are a widely applied classifier. In this talk I present the recent bounds by Bartlett et. on the VC dimension of a network employing the popular rectified (or ReLU) activation function. For a network of L layers and W weights they prove the almost tight bounds Omega(WLlog(W/L) and O(WLlog(W). I'll further talk about recent efforts in characterizing the induced input space partition of a ReLU network and hint at possible research directions.

Upcoming talks     |     All previous talks     |     Talks by speaker     |     Upcoming talks in iCal format (beta version!)

Previous talks by year:   2024  2023  2022  2021  2020  2019  2018  2017  2016  2015  2014  2013  2012  2011  2010  2009  2008  2007  2006  2005  2004  2003  2002  2001  2000  1999  1998  1997  1996  

Information for students and suggested topics for student talks

Automatic MiSe System Software Version 1.4803M   |   admin login