Post-Doctoral Research Visit F/M Graybox Tunneling for Multi-Objective Combinatorial Optimization: Algorithmic and HPC Perspectives

Inria Villeneuve d'Ascq; Villeneuve-d'Ascq; On-site; EUR 36,000 - 48,000; Full time

Job summary

1000scholars seeks a postdoctoral researcher to design, develop, and validate graybox tunneling algorithms for multi-objective combinatorial optimization, focusing on parallelization on HPC platforms. The researcher will extend single-objective tunneling to the multi-objective setting and develop strategies for exploring the Pareto front.

The role includes investigating parallelism at solution, iteration, and algorithm levels, exploring island-based hybrids, and implementing scalable software

Qualifications

• Strong background in combinatorial optimization, evolutionary computation, computational intelligence, or related areas.
• Experience with parallel and HPC computing is advantageous.

Responsibilities

• Design graybox tunneling algorithms for multi-objective optimization exploiting problem structure.
• Develop parallel tunneling strategies to exploit multiple sources of parallelism.
• Design hybrid island-based optimization to concurrently explore Pareto front regions.
• Implement and analyze scalability of tunneling algorithms on HPC platforms.

Skills

Combinatorial optimization

Evolutionary computation

Computational intelligence

Heuristic search

Parallel HPC computing

Job description

Assignment

The postdoctoral researcher will contribute to the design, development, and experimental validation of novel graybox tunneling algorithms for multi-objective combinatorial optimization, with a particular focus on their parallelization and scalability on high-performance computing (HPC) platforms. The main assignment will be to investigate how knowledge about the structure and local interactions of combinatorial optimization problems can be exploited to design efficient tunneling mechanisms capable of navigating large sets of Pareto local optima. In particular, the researcher will contribute to extending existing single-objective tunneling principles to the multi-objective setting and to developing new algorithmic strategies for efficiently exploring different regions of the Pareto front. A second assignment will concern the identification and exploitation of different levels of parallelism arising within tunneling-based multi-objective search. This can include solution-level and iteration-level parallelism, as well as higher-level parallelism based on concurrent tunneling processes. For exemple, the researcher will investigate hybrid optimization strategies, notably based on distributed island models, in which different tunneling processes explore complementary regions of the search space and exchange selected Pareto local optimal solutions. The researcher will also contribute to the implementation and empirical evaluation of the resulting algorithms on modern HPC platforms. The position is open to candidates with a strong background in combinatorial optimization, evolutionary computation, computational intelligence, heuristic search, or related areas, as well as to candidates with expertise in parallel and HPC computing combined with a solid understanding of combinatorial optimization. While prior HPC experience is an advantage for candidates from optimization and computational intelligence backgrounds, candidates primarily specialized in HPC are also encouraged to apply provided they have relevant expertise in combinatorial optimization. The work will ultimately contribute to the development of prototype software libraries for parallel tunneling and to systematic computational studies demonstrating their effectiveness and scalability on multi-objective combinatorial optimization problems, such as, NK-landscapes, SAT, TSP.

Main activities

• Design of graybox tunneling algorithms for multi-objective combinatorial optimization, exploiting structural information about problem variables and their interactions to efficiently navigate sets of Pareto local optima.
• Development of parallel tunneling strategies, identifying and exploiting solution-level, iteration-level, and algorithm-level sources of parallelism to enhance the exploration of large combinatorial search spaces.
• Design of hybrid island-based optimization approaches, enabling multiple tunneling processes to concurrently explore complementary regions of the Pareto front and exchange selected Pareto local optima.
• Implementation and scalability analysis of parallel tunneling algorithms on HPC platforms, initially targeting multi-core and multi-CPU architectures and possibly investigating the use of many-core architectures where appropriate.
• Experimental validation and comparative evaluation of the proposed methods on multi-objective combinatorial optimization problems.

The additional activities of the postdoctoral researcher will include:

• Development of reusable software for parallel graybox tunneling, supporting large-scale and reproducible computational experiments.
• Scientific dissemination, including preparation of research papers, conference presentations, technical reports, etc
• Research collaboration and knowledge exchange with researchers in combinatorial optimization, evolutionary computation, computational intelligence, and high-performance computing.

Skills

The position is open to candidates with a strong background in combinatorial optimization, evolutionary computation, computational intelligence, heuristic search, or related areas, as well as to candidates with expertise in parallel and HPC computing combined with a solid understanding of combinatorial optimization. While prior HPC experience is an advantage for candidates from optimization and computational intelligence backgrounds, candidates primarily specialized in HPC are also encouraged to apply provided they have relevant expertise in combinatorial optimization.