The Evolution of Hybrid Soft Computing: A Personal Journey

This paper presents the author's personal journey throughout the inception of Soft Computing (SC) and its evolution into Hybrid Soft Computing (HSC). In 1994, Zadeh de ned the concept of Soft Computing as a loose association or partnership of components . Since then, SC has undergone several transformational phases. In our journey, we provide a retrospective view of SC past four phases, its current view ( fth phase) and a prospective view, in which we envision the role of SC within the new trend of model ensembles and model fusion. The retrospective view starts with the hybridization phase, which is driven by the inherit ease of integration of SC components and leads to the concept of Hybrid Soft Computing. The second phase, inspired by traditional AI problem formulations, is a two-level model characterization based on object-level and meta-level reasoning. The third phase addresses the knowledge and meta-knowledge representation required by each of these reasoning levels using a linguistics analogy. The fourth phase is an extension of the heuristics used at the meta-level, i.e., Meta-heuristics (MH's), from evolutionary algorithms to other global search methods. The current view ( fth phase) consolidates the above phases and separates o ine MH's, used for design and tuning, from online MH's, used for run-time selection or aggregation of object-level models. This view suggests a broader use of SC components, since it enables us to use hybrid SC techniques at each of the MH's levels as well as at the object level. Furthermore, this separation facilitates model lifecycle management, which is required to maintain the models' vitality and prevent their obsolescence over time. Finally, the prospective view analyzes the recent trend of using model ensembles rather than individual models: the elements of the ensemble are the object-level models, while the fusion mechanisms are the online MH's that select or aggregate them.