The Promise of Unifying Metaheuristics and Analytics
Meta-Analytics represents the unification of metaheuristics and analytics, two fields of the foremost interest and practical importance for applications ranging from biotechnology to energy to logistics
and financial planning. While metaheuristics provide a modern framework and an arsenal of cutting-edge techniques to handle complex, real-world problems, Analytics embodies the use of machine learning and optimization in practical contexts.
Thus, their marriage can be regarded as a natural step towards both the creation of effective tools for problems in the Analytics domain and the expansion of the scope of metaheuristic techniques.
An important development in Meta-Analytics has come about with the emergence of Quantum Bridge Analytics, a field devoted to bridging the gap between classical and quantum computational methods and technologies.
As emphasized in the 2019 Consensus Study Report titled Quantum Computing: Progress and Prospects, by the National Academies of Sciences, Engineering and Medicine quantum computing will remain in its infancy for another 5 to 10 years, and in the interim “formulating an R&D program with the aim of developing commercial applications for near-term quantum computing is critical to the health of the field." As noted in this report, such a program will rest on developing “hybrid classical-quantum techniques,” which is the focus of Quantum Bridge Analytics. This branch of Meta-Analytics is being actively pursued with the creation of the Alpha-QUBO solver, whose forerunner has been embodied in a hybrid classical-quantum method called qbsolv applied in a wide range of commercial and academic research settings.
The following sections describe the advantages obtained by the synergies of the techniques and the avenues for achieving such a unification of methodologies provided by Meta-Analytics and discusses some important themes in the field.
The Meta-Analytics theme has its origins in a series of seminal developments in optimization and machine learning and their practical applications. The term “Analytics” has gained recognition as a referent for analyses that embody prediction and optimization in a broad sense, typically supported by interpretive aids for users. As a result, organizations from a wide range of disciplines have allied themselves with the Analytics area. This includes researchers and practitioners in classical optimization, notably in the fields of engineering, computer science, operations research and management science. As a compelling example, the prestigious Institute of Management Science and Operations Research (INFORMS) has adopted the area of Analytics as a primary focus, and has created a new magazine called Analytics.
Metaheuristics (Glover, 1986; Blum and Roli, 2003; Glover and Kochenberger, 2003; Sörensen et al, 2017), emerged from the dawning recognition that many real world problems in business, science and industry are too large or too
complex for classical optimization methods to handle effectively. To remedy this problem, recourse was initially made to
joining classical optimization with simple heuristic methods, but it soon became clear that more powerful approaches were
needed, incorporating various ideas of heuristics, but going beyond them. The methods of metaheuristics were conceived to
meet this challenge, with innovative evolutionary and neighborhood search approaches whose first forms appeared in the late
1960s and early 1970s, and which have since undergone substantial refinements and modifications. The name “metaheuristics”
itself emerged in the mid-1980s, as the recognition of the essential focus of these new methods became universal.
Meta-Analytics represents the unification of Metaheuristics and Analytics. While the former provide
a modern framework and an arsenal of cutting-edge techniques to handle complex, real-world problems, the latter embodies the
use of prediction and optimization techniques in practical contexts. Thus, their marriage can be only regarded as a natural step
towards both the creation of effective tools for problems in the Analytics domain and the expansion of the scope of
metaheuristic techniques. Indeed, modern versions of these latter techniques, such as evolutionary algorithms (Eiben and
Smith, 2003), tabu search (Glover and Laguna, 1997), simulated annealing (Dekkers and Aarts, 1991), swarm intelligence
algorithms (Kennedy and Eberhart, 2001), memetic algorithms (Neri et al, 2012; Cotta et al, 2016) and a variety of others
which have proved to be highly successful, producing an explosion of publications in international journals and presentations
at international conferences. These developments have additionally resulted in the formation of new journals and new societies.
The ability to deal with challenging practical problems more effectively, including those from domains that Analytics claims as
its focus, lays a foundation for an alliance between Metaheuristics and Analytics. This is notably exemplified by the fact that
the Metaheuristics field has made important contributions to predictive and prescriptive analysis, which are prominent concerns
of Analytics. The emergence of Quantum Bridge Analytics within Meta-Analytics has further brought these developments together with quantum computing and machine learning.
The unification of Metaheuristics and Analytics within Meta-Analytics brings about important advantages that were not
fully realized in the past as these two fields evolved largely in isolation from each other. Chief among these are the promise of
Metaheuristics to become a source of more effective tools for problems in the Analytics domain, and in turn the promise of
Analytics to provide a perspective for expanding the scope of algorithmic methods within Metaheuristics. These potentials are
accentuated by the fact that many researchers and practitioners in Analytics have not been exposed to the Metaheuristics field,
and are unaware of its power for addressing practical applications, while many of those working within Metaheuristics have
incompletely appreciated the value of incorporating elements that have become the purview of Analytics. The establishment of
Meta-Analytics creates an opportunity to reach an expanded community of decision makers in industry, science and
government who can profit from the union of its component areas.
The scope of this union can be glimpsed by elucidating the primary themes of Meta-Analytics and by looking at some previous approaches that have been paving the way. Subsequently, we shall describe the advantages obtained by the synergies of the techniques and the avenues for achieving such a unification of methodologies. We also introduce contributions contained in this section, in which these themes are explored in more detail.
Themes of Meta-Analytics
Six themes broadly constitute the main thrusts of the Meta-Analytics area:
1) Using Metaheuristics as a source for creating enhanced predictive and machine learning methods (as in clustering, discrimination, feature detection, pattern recognition and classification, etc.). While such concerns have long been a part of the metaheuristic domain, a more dedicated emphasis on them through Meta-Analytics lays a foundation for significant new advances. A key source of contributions derives from highlighting such advances for their general relevance to Analytics, and hence to the Metaheuristics/Analytics union.
2) Incorporating predictive and machine learning methods to enhance the performance of metaheuristics. In spite of a variety of proposals for exploiting learning within metaheuristics, e.g., see (Glover and Greenberg, 1989; Kelly et al,1996; Birattari, 2009), very little has been done to pursue this theme. Predictive and machine learning procedures can
be applied with metaheuristics in offline (pre-solution and post-solution) stages as well as during run time execution. An important step forward will be supplied by refining and implementing meritorious ideas which have been inadequately investigated, and by developing new proposals to capitalize on the opportunities opened up by joining analytics and metaheuristics.
3) Creating special mechanisms and interfaces for interpreting outcomes and relationships uncovered by metaheuristic solution processes. This focus has the goal of enabling users to interact with Meta-Analytic procedures to achieve greater insight and yield better decisions. This interaction includes adaptive exploration of model assumptions as well as decision rules for guiding the methods studied (Meignan et al, 2015).
4) Developing integrative methods that capitalize on one or more of the preceding themes to build highly effective algorithms that utilize domain knowledge for solving problem from important classes. Tabu search (Glover and Laguna, 1997) and Memetic algorithms (Neri et al, 2012; Cotta et al, 2016) are good examples of this theme.
5) Creating improved methods for analyzing and explaining the operation of alternative solution approaches, including more effective and comprehensive forms of landscape analysis and quasi-decomposition analysis as embodied in vocabulary building strategies.
6) Establishing a repository of important applications in business, science and government where Meta-Analytics provides advances of singular value, leading to improved insights, operations and policies.