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

Date and Time: Thursday, March 17, 2022, 12:15 pm

Duration: 30 minutes

Location: OAT S15/S16/S17

Speaker: Simon Weber

Training Fully-Connected Neural Networks is ER-Complete

We consider the algorithmic problem of finding the optimal weights to a two-layer ReLU fully-connected neural network, to fit a given set of data points. We show that the problem is ∃R-complete. Our results hold even if the following restrictions are all added simultaneously. •There are exactly two output neurons. •There are exactly two input neurons. •The data has only a constant number of different labels. •The number of middle neurons is linear in the number of data points. •The data needs to be fitted exactly. This shows that even the simplest possible neural networks are difficult to train. This is joint work with Daniel Bertschinger, Christoph Hertrich (London), Paul Jungeblut (Karlsruhe), and Tillmann Miltzow (Utrecht).

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