Low-loss connection of weight vectors: distribution-based approaches

08/03/2020
by   Ivan Anokhin, et al.
0

Recent research shows that sublevel sets of the loss surfaces of overparameterized networks are connected, exactly or approximately. We describe and compare experimentally a panel of methods used to connect two low-loss points by a low-loss curve on this surface. Our methods vary in accuracy and complexity. Most of our methods are based on "macroscopic" distributional assumptions, and some are insensitive to the detailed properties of the points to be connected. Some methods require a prior training of a "global connection model" which can then be applied to any pair of points. The accuracy of the method generally correlates with its complexity and sensitivity to the endpoint detail.

READ FULL TEXT

page 1

page 2

page 3

page 4

research
01/22/2019

On Connected Sublevel Sets in Deep Learning

We study sublevel sets of the loss function in training deep neural netw...
research
09/15/2020

Evaluating representations by the complexity of learning low-loss predictors

We consider the problem of evaluating representations of data for use in...
research
11/30/2014

The Loss Surfaces of Multilayer Networks

We study the connection between the highly non-convex loss function of a...
research
03/22/2005

Analytic Definition of Curves and Surfaces by Parabolic Blending

A procedure for interpolating between specified points of a curve or sur...
research
03/03/2022

Counting points on abelian surfaces over finite fields with Elkies's method

We generalize Elkies's method, an essential ingredient in the SEA algori...
research
06/14/2022

Prioritized Training on Points that are Learnable, Worth Learning, and Not Yet Learnt

Training on web-scale data can take months. But most computation and tim...
research
03/07/2019

Locating Transparent Objects to Millimetre Accuracy

Transparent surfaces, such as glass, transmit most of the visible light ...

Please sign up or login with your details

Forgot password? Click here to reset