Toward a Standarised Description of Optimization Techniques

In considering the insane notion to implement all computational intelligence techniques, one can't help but think about standardisation, simply as a method for reducing effort. A particular standardisation I have been using in my thesis, and talking over with D is that of the description of biologically inspired optimisation algorithms, that naturally extends to all of CI.

The idea was born out of the question as to how to communicate an approach in one page. Specifically, the salient features and capabilities of an approach (also valid for a class of approaches) would be distilled into a single A4 page, along the lines of a cheat sheet or a one page resume. The reason for such compression is not to communicate the in's and out's of a given approach (there are plenty of works out there that do that well enough), instead, the idea was proposed out of the necessity to compress the information on approaches to effectively communicate choice between approaches.

The general framework involved describing each approach in terms of four key perspectives:

• Metaphor: The inspiration that motivated the human in the development of the approach, providing a common context for the strategy.
  
• Strategy: The information processing properties and emergent behaviours of the approach, describing the principles and expectations for the operations.
  
• Operations: A technical summation of the bottom-up procedures for achieving the strategy, sufficient for specialisation and implementation.
  
• Further Reading: Lists primary sources and seminal works supporting and elaborating the compressed descriptions.
  

The first three perspectives were chosen as they were identified as the three most common interest in contrasting and comparing approaches (besides results). Descriptions are generalisations of problem and implementation specific realisations. The audience is the general practitioner. Ideally, Wikipedia (or similar Wiki) would achieve such descriptions, and use a portal to unify perspectives on the art (fields of study).

The collection and communication of the approaches in this method would provide insights into both important similarities and differences between approaches, potentially promoting improved hybridisation, informed extension, and at the very least a broader perspective than that of the Computational Intelligence or Metaheuristics perspectives (or subsidiaries).

The vision for presenting this work is either a book (Algorithm Atlas) and a web page, potentially hooked into the mapping of NFL, and potentially generally editable toward community-based improvement.