Why LASSO, EN, and CLOT

Invariance-Based Explanation

authored by
Hamza Alkhatib, Ingo Neumann, Vladik Kreinovich, Chon Van Le
Abstract

In many practical situations, observations and measurement results are consistent with many different models—i.e., the corresponding problem is ill-posed. In such situations, a reasonable idea is to take into account that the values of the corresponding parameters should not be too large; this idea is known as regularization. Several different regularization techniques have been proposed; empirically the most successful are LASSO method, when we bound the sum of absolute values of the parameters, and EN and CLOT methods in which this sum is combined with the sum of the squares. In this paper, we explain the empirical success of these methods by showing that they are the only ones which are invariant with respect to natural transformations—like scaling which corresponds to selecting a different measuring unit.

Organisation(s)
Geodetic Institute
External Organisation(s)
University of Texas at El Paso
Vietnam National University Ho Chi Minh City
Type
Contribution to book/anthology
Pages
37-50
No. of pages
14
Publication date
14.11.2020
Publication status
Published
Peer reviewed
Yes
ASJC Scopus subject areas
Artificial Intelligence
Electronic version(s)
https://digitalcommons.utep.edu/cgi/viewcontent.cgi?article=2350&context=cs_techrep (Access: Open)
https://doi.org/10.1007/978-3-030-48853-6_2 (Access: Closed)
 

Details in the research portal "Research@Leibniz University"