[Ml-stat-talks] Fwd: Next PACM Colloquium - Andrew Barron, Yale University
ehazan at CS.Princeton.EDU
Wed Feb 17 21:25:12 EST 2016
Talk of interest next week.
---------- Forwarded message ----------
From: Liz Van Dzura <ldzura at princeton.edu>
Date: Tue, Feb 16, 2016 at 9:56 AM
Subject: Next PACM Colloquium - Andrew Barron, Yale University
To: Liz Van Dzura <ldzura at princeton.edu>
The next PACM Colloquium will be presented by: Andrew Barron, Yale
Monday, February 22, 4:30 pm – Fine Hall 214
TITLE: Computationally feasible greedy algorithms for neural nets
ABSTRACT: Previously, greedy algorithms have been shown to have good
approximation and estimation properties for superpositions of a sigmoidal
function or other ridge activation functions. In each step the parameters
of a new sigmoid are fit to the residuals of the previous sigmoids.
However, that has remained a challenging task to produce guarantees for the
parameter search for each sigmoid.
Here we discuss developments of two algorithms for this task. One is an
implementation of adaptive annealing, in which internal modifications of
the sigmoid is made, including parameter squashing and variable replication
that permit individual parameter effects to be small, allowing stability of
the adaptive annealing solution, at the expense of increasing
dimensionality. The other algorithm is a convergent nonlinear modification
of the tensor methods of Anandkumar and her colleagues, motivated by
optimization of the inner product of the residuals and certain forms of
sigmoids, in the context of data from Gaussian or other rotationally
invariant distributions. This work is joint with Jason Klusowski.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Ml-stat-talks