Daniel
3efd8a6c26
Some checks are pending
CI / Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }} (x64, ubuntu-latest, 1.10) (push) Waiting to run
CI / Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }} (x64, ubuntu-latest, 1.6) (push) Waiting to run
CI / Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }} (x64, ubuntu-latest, pre) (push) Waiting to run
12 lines
2.1 KiB
TeX
12 lines
2.1 KiB
TeX
\chapter{Abstract}
|
|
|
|
The objective of symbolic regression is to identify an expression that accurately models a system based on a set of inputs. For instance, one might determine the flow through pipes using inputs such as roughness, diameter, and length by conducting experiments with varying input configurations and observing the resulting flow and derive an expression from the experiments. This methodology, exemplified by \textcite{nikuradse_laws_1950}, can be applied to any system through symbolic regression. To find the best-fitting expression, millions of candidate expressions are generated, each requiring evaluation against every input configuration to assess how well they fit to the system. Consequently, millions of evaluations must be performed, a process that is computationally intensive and time-consuming. Thus, optimizing the evaluation phase of symbolic regression is crucial for discovering expressions that describe large and complex systems within a feasible timeframe.
|
|
|
|
% Applications such as weather simulation \parencite{michalakes_gpu_2008}, simulation of static and rotating black holes \parencite{hissbach_overview_2022, verbraeck_interactive_2021}, and structural analysis \parencite{georgescu_gpu_2013} significantly benefit from optimized algorithms that leverage the graphics processing unit (GPU).
|
|
|
|
This thesis presents the design and implementation of two evaluators that utilize the GPU to evaluate expressions generated at runtime by the symbolic regression algorithm. Performance benchmarks are conducted to compare the efficiency of the GPU evaluators against the current CPU evaluator.
|
|
|
|
The benchmark results indicate that the GPU can serve as a viable alternative to the CPU in certain scenarios. The determining factor for choosing between GPU and CPU evaluation is the number of input configurations. In a scenario with $10\,000$ expressions and $10\,000$ input configurations, the GPU outperformed the CPU by a significant margin.
|
|
|
|
This master thesis is associated with the FFG COMET project ProMetHeus (\#904919). The developed software is used and further developed for modelling in the ProMetHeus project.
|