I am a fourth year Ph.D student and have been with the lab since its inception. I am developing knowledge-intensive multi-objective optimization methods by integrating domain knowledge with multiobjective evolutionary algorithms (MOEA) and adaptive operator selection (AOS). Domain-knowledge such as expert knowledge or lessons learned are expected to better guide a conventional MOEA in its search process and increase its efficiency. However, not all domain-knowledge helps improve the search, so an AOS is used to apply both knowledge-dependent operators alongside knowledge-independent operators (e.g. mutation and crossover) with the goal of focusing computational resources on the most effective operators. This not only increases the computational efficiency of the MOEA, but also boosts its robustness in its ability to solve a given problem because it can adapt its search strategy on-the-fly, which increases the MOEA’s ability to escape local optima if the search stagnates.

contact: nh295(@)

NSTRF 2016 Research

My research on knowledge-intensive optimization methods is supported by a NASA Science and Technology Research Fellowship (NSTRF 2016). The abstract for the research can be found here (abstract link). Specifically, I will use knowledge-intensive MOEAS to support decision-making in designing distributed satellite missions (DSM). The space of possible DSM designs is extremely large as a result of many decision variables including the orbital parameters, payload, and bus components for each satellite. Moreover, these highly coupled design variables affect figures of merit (e.g. revisit time, spatial resolution) in non-obvious ways. Therefore, efficient methods are necessary to not only search the vast space of system architectures in order to find acceptable solutions, but also learn the different tradeoffs in the figures of merit.

Another component of my NASA supported research is to apply data mining methods to the intermediate or final results of an optimization run in order to discover the driving design decisions that lead to good designs. These driving design decisions not only help to explain the results, but also can be be fed back into the MOEA as new operators to try to improve the algorithm’s efficiency further.



I have also supported the tradespace analysis for the Time-Resolved Observations of Precipitation structure and storm Intensity with a Constellation of Smallsats (TROPICS) mission (main website). I investigated different constellation configurations by varying the number of satellites in the constellation and their orbital parameters. In a preliminary study, I also examined how the payload affected the cost (see AGU poster below). As the mission requirements are developed, I will continue to support architecture design with the tools developed above through the NSTRF research.


Journal papers:

Hitomi, N., Selva, D. “A Classification and Comparison of Credit Assignment Strategies in Multiobjective Adaptive Operator Selection”. IEEE Transactions on Evolutionary Computation, vol.PP, no.99, pp.1-1
doi: 10.1109/TEVC.2016.2602348 [in Print]

Conference papers:

Hitomi, N., and Selva, D. 2016. “Hyperheuristic approach to leveraging domain knowledge in multiobjective evolutionary algorithms.” In ASME 2016 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference.

Hitomi, N., and Selva, D. 2015. “Experiments with Human Integration in Asynchronous and Sequential Multi-Agent Frameworks for Architecture Optimization.” Procedia Computer Science 44: 393–402.

Selva, D., Abello C., and Hitomi N. 2015. “Preliminary Experiments with Learning Agents in an Interactive Multi-Agent Systems Architecture Tradespace Exploration Tool.” In 2015 Annual IEEE Systems Conference Proceedings, 445–452.

Hitomi, N., and Selva, D. 2015. “The Effect of Credit Definition and Aggregation Strategies on Multi-Objective Hyper-Heuristics.” In ASME 2015 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference.


Hitomi, N., Selva, D. 2016. “Adaptive Domain-Specific Heuristics for Optimizing Systems Architecture”. CESUN 5th International Engineering Systems Symposium.

Hitomi, N., Selva, D., and Blackwell, W. 2015. “Exploring the Architectural Tradespace of Severe Weather Monitoring Nanosatellite Constellations.” American Geophysical Union Fall Meeting 2015. (link to poster)


Multi-objective Credit Assignments for Adaptive Operator Selectors (also known as Hyperheuristics)

An adaptive operator selector (AOS) is a multi-method strategy that adaptively selects high performing operators (e.g. mutation, crossover) to solve the problem at hand. The purpose of AOS is twofold. The first is to elevate the generality of an optimization method by coordinating cooperation between operators such that the weakness of one operator is covered by the strengths of other. The second is to allocate computational resources to favor high performing operators for increased efficiency.

Credit-assignment is the method that rewards operators for its impact on the search process, but currently, we lack an understanding of effective credit-assignment methods on a multiobjective problem. My work published in IEEE Transactions on Evolutionary Computation (link to paper) focuses on identifying effective performance metrics for operators on multiobjective problems.