Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Code and Circuit Generation for Efficient Projection Computations on Embedded Platforms


S. Merkli

Master Thesis,HS13 (10308)

Minimum Euclidian norm projection onto polyhedral sets is a core operation in first-order methods for convex optimization problems with convex affine inequality constraints. This projection is performed at every iteration of the solver, which makes its efficiency critical for the overall performance. For many practical problems in embedded optimization, this projection can be explicitly written as an evaluation of a piecewise affine function. State-of-the-art architectures evaluate such functions by iteratively traversing binary trees. This thesis describes a novel approach that uses graph representations of the computations to be performed to optimize operation sharing. Automatic circuit generation is used to obtain problem-specific implementations that can significantly outperform current reference implementations with only modest increases in circuit size. The results are implemented in a software toolchain and quantitative comparisons with existing toolchains are made, demonstrating the effectiveness of the approach.


Type of Publication:

(12)Diploma/Master Thesis

J.L. Jerez

File Download:

Request a copy of this publication.
(Uses JavaScript)
% Autogenerated BibTeX entry
@PhdThesis { Xxx:2014:IFA_4755
Permanent link