Experiments in genetic divergence for emergent systems

McGowan, Christopher; Wild, Alexander; Porter, B.

doi:10.1145/3194810.3194813

Cited by 4 publications

(7 citation statements)

References 22 publications

(21 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The second research direction is to consider additional computational problems associated with emergent software systems. Of particular interest here is the automated synthesis of both variants of existing components and new components from input-output examples [47,48]. Analyses of these problems may also give insights into the more general problem of program synthesis [49,50,51] Observe that interfaces base, vertStatJ for 1 ≤ J ≤ |V |, and domSetStat and components VertexStatusJ for 1 ≤ J ≤ |V |, DomSetStatus0, and DomSetStatus1 are the same as in the reduction in the proof of Result A.9.…”

Section: Discussionmentioning

confidence: 99%

Viable Algorithmic Options for Creating and Adapting Emergent Software Systems

Wareham¹,

Haan²

2022

Preprint

View full text Add to dashboard Cite

Given the complexity of modern software systems, it is of great importance that such systems be able to autonomously modify themselves, i.e., self-adapt, with minimal human supervision. It is critical that this adaptation both results in reliable systems and scales reasonably in required memory and runtime to non-trivial systems. In this paper, we apply computational complexity analysis to evaluate algorithmic options for the reliable creation and adaptation of emergent software systems relative to several popular types of exact and approximate efficient solvability. We show that neither problem is solvable for all inputs when no restrictions are placed on software system structure. This intractability continues to hold relative to all examined types of efficient exact and approximate solvability when software systems are restricted to run (and hence can be verified against system requirements) in polynomial time. Moreover, both of our problems when so restricted remain intractable under a variety of additional restrictions on software system structure, both individually and in many combinations. That being said, we also give sets of additional restrictions that do yield tractability for both problems, as well as circumstantial evidence that emergent software system adaptation is computationally easier than emergent software system creation.(and hopefully more manageable) subsystems [14,15]. It is precisely at this current stage of new algorithm development that computational complexity analyses might be most useful. Previous WorkComputational complexity analyses have been done previously for component-based software system creation by component selection [16,17] and component selection with adaptation [18], with [18,17] having the additional requirement that the number of components in the resulting software system be minimized. Given the intractability of all of these problems, subsequent work has focused on efficient approximation algorithms for component selection. Though it has been shown that efficient algorithms that produce software systems whose number of components is within a constant multiplicative factor of optimal are not possible in general [19], efficient approximation algorithms are known for a number of special cases [20,21,19]. All of these analyses assume that any component can be composed with any other component, in the sense that function definitions not given in a component c can be obtained by composition of c with any other components that have the required function definitions, i.e., component composition is not regulated using component interfaces. Moreover, none of the formalizations used include specification of the internal structure of software systems, system requirements, and components that are detailed enough to allow investigation of restrictions on these aspects that could make component selection or component selection with adaptation tractable.Only one analysis to date has incorporated a component model which allows investigation of the tractability effects of restrictions ...

show abstract

Section: Discussionmentioning

confidence: 99%

Viable Algorithmic Options for Creating and Adapting Emergent Software Systems

Wareham¹,

Haan²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…The offline code synthesiser, however, is the novel part of the proposed framework. Inspired by recent advances in code synthesis with techniques ranging from genetic improvement [4] to neural networks [5], we envision the code synthesiser as a key element to advance self-improving systems' state-of-the-art.…”

Section: Approachmentioning

confidence: 99%

“…Depending on the component, different code synthesis strategies should be applied. For example, for components that are generated from improving existing components, such as cache components, which are generated by improving their hash function, techniques such as genetic improvement has been shown effective [4], whereas the synthesis of entirely new components may use neural networks as described in [5].…”

Section: Approachmentioning

confidence: 99%

Code Synthesis in Self-Improving Software Systems

Filho

Wild

Porter

2019

2019 IEEE 4th International Workshops on Foundations and Applications of Self* Systems (FAS*W)

Self Cite

View full text Add to dashboard Cite

This paper aims at provoking a wide discussion around code synthesis and its importance in creating the next generation of self-improving software systems. As a starting point for the discussion, a machine-to-machine self-improving framework is presented. The framework aims at improving system's performance by integrating two modules: i) a selfimproving online module, and ii) a code synthesiser offline module. The online module learns, at runtime, as it handles the system's inputs, how to best compose the system from a pallet of available software components and a user-defined high level goal. The offline code synthesiser generates new components based on the perceived system's input, executing environment and the system's goal provided by the online module. The code synthesiser then provides better component options for the online module to integrate with the system to improve its performance. The paper describes the framework, focusing on its main challenges.

show abstract

“…These systems rely on the existence of variation in their pool of building blocks to learn better compositions of behaviour -such as different sorting algorithms, cache replacement algorithms, or load balancing policies. While most systems require human engineers to generate this pool of variation, an initial study by McGowan et al indicated that genetic improvement (GI) could automate the generation of environment-tailored variation with the example of a hash table [3]. Based on this study, we examine the wider set of challenges in genetic improvement for emergent systems and some of the most promising research directions to explore.…”

Section: Introductionmentioning

confidence: 98%

“…b) Generalisation: While optimisation to a given target is important, genetic improvement algorithms typically aim towards a fixed environment -for example optimising a particular component towards a specific set of inputs. Rather than using GI to derive a hash function for all words in the English language, we would instead use a smaller subset of commonly-seen inputs from a particular environment [3]. However, this presents a potential dichotomy between spe-[pre-print version for personal use] Appears at IEEE GI 2021 cialising on a smaller set of inputs or generalising on a much broader set -a problem analogous to over-fitting in wider machine learning and a significant problem in all evolutionary systems.…”

Section: Introductionmentioning

confidence: 99%