Department of Computer Science | Institute of Theoretical Computer Science | CADMO

Prof. Emo Welzl and Prof. Bernd Gärtner

Mittagsseminar Talk Information |

**Date and Time**: Tuesday, October 28, 2014, 12:15 pm

**Duration**: 30 minutes

**Location**: OAT S15/S16/S17

**Speaker**: Hemant Tyagi

We consider the problem of learning Sparse Additive Models (SPAMs), i.e. functions of the form: $f(x_1,...,x_d) = \sum_{l \in S} \phi_{l}(x_l)$; $S \subset {1,\dots,d}$, from point queries of $f$. Here $\phi_l$ and $S$ are unknown and $|S| = k \ll d$. Such models have been studied extensively in statistics, in the context of non-parametric regression. However in that setting, one typically does not have control over the points, at which the value of $f$ is observed.

Assuming $\phi_l$'s to be smooth, we propose a set of points at which to query $f$ and an efficient randomized algorithm that recovers: (i) $S$ exactly and, (ii) a uniform approximation to each $\phi_l$; for all $l \in S$. In contrast, the existing results in statistics provide error bounds for recovering $f$ in the weaker mean square sense.

If the point queries are noiseless, we show that $O(k \log d)$ queries suffice while if the point queries are corrupted with (i.i.d) Gaussian noise, we show that $O(k^3 (\log d)^2)$ queries suffice.

Joint work with Bernd Gärtner and Andreas Krause.

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