Dispersive Flies Optimisation¶
A bounded population optimizer that moves coordinates from a fitter ring neighbor using the current swarm-best displacement, with thresholded disturbance.
Core Idea¶
Dispersive Flies Optimisation (DFO) searches a bounded numerical space using a population of candidate positions called flies. The original method scores each candidate with a quantity to minimize. For each coordinate, the better of a fly's two ring neighbors supplies a starting point, and a randomly scaled difference between the current swarm best and the current fly supplies a move. A threshold condition can replace that coordinate with a bound-based value. This is a computational rule inspired by swarming, not a model of real fly behavior.[^ref-89d9a4e9f710]
The original algorithm does not state that the best fly is kept unchanged. It updates every fly in its loop. It also does not keep the velocity or personal-best memory used by the compared Particle Swarm Optimization method.[^ref-89d9a4e9f710]
Scope of Application¶
DFO applies when a problem can be written as bounded numerical decision variables and a fitness function that ranks candidates. The original paper used continuous benchmark functions. A later paper reports using three fly coordinates for PID controller gains in a simulated DC motor, minimizing integral time-weighted absolute error (ITAE). The second paper reports a model and simulation, not a physical motor deployment or independently inspected implementation code.[ref-89d9a4e9f710][ref-1b4bd6e2f478]
Clarity¶
“Best” means the current lowest-fitness member of the population; “better neighbor” means the lower-fitness fly of the two ring neighbors. The neighbor supplies the proposal's center, while the swarm best supplies a displacement relative to the current fly. The original disturbance line tests \(r<d_t\) and then uses the same printed \(r\) in \(x_{\min,d}+r(x_{\max,d}-x_{\min,d})\). Because the paper does not state that a fresh draw occurs after the test, the conditional replacement should not be called uniformly distributed over the full bounds.[^ref-89d9a4e9f710]
Manages Complexity¶
The method packages a numerical search into candidate bounds, a fitness function, population size, disturbance threshold and stopping budget. It uses scored members of its own population to make new proposals, without asking for a derivative of the objective. That simplicity still leaves evaluation cost and parameter choice to the problem at hand. The original benchmark comparisons do not prove that DFO will win on every landscape.[^ref-89d9a4e9f710]
Abstract Reasoning¶
DFO separates two forms of guidance. One is local to the ring: choose a better nearby fly. The other uses the population's current best position to determine a coordinate displacement. The optional disturbance can interrupt that guided move. When studying a run, ask whether the guides are useful for the given objective and whether disturbance finds better candidates often enough to justify the guided moves it replaces.[^ref-89d9a4e9f710]
Knowledge Transfer¶
Before applying a paper's “DFO” result elsewhere, identify what each coordinate represents, what the fitness function minimizes, the bounds, the exact update formula, and the evaluation budget. A method with a changed differential may still be named DFO, but its results do not directly establish the behavior of the original rule. The 2016 medical-imaging study, for example, changes the differential from swarm-best-minus-current to neighbor-best-minus-current.[ref-89d9a4e9f710][ref-ee2c81317899]
Example¶
In the 2021 motor-control simulation, each fly encodes proportional, integral and derivative gains \(K_p,K_i,K_d\), each bounded between 0.01 and 20. The authors minimize ITAE, print the baseline ring-neighbor and swarm-best coordinate move, and report 50 flies, a disturbance threshold of 0.001 and 100-iteration comparisons. Its pseudocode also assigns \(x_k\) within an \(i\)-loop and prints an evaluation cap inconsistent with that 100-iteration table. The example therefore shows a source-reported application, not proof of exact code behavior.[^ref-1b4bd6e2f478]
Relationships to Other Abstractions¶
Current abstraction Dispersive Flies Optimisation Domain-specific
Parents (2) — more general patterns this builds on
-
Dispersive Flies Optimisation is a kind of Algorithm Prime
DFO is a finite randomized input-to-output procedure with prescribed update steps and a stopping budget.
-
Dispersive Flies Optimisation presupposes Optimization Problem Domain-specific
DFO requires bounded decision vectors, a fitness objective, and a minimization sense to select its guides.
Hierarchy paths (3) — routes to 3 parentless roots
- Dispersive Flies Optimisation → Algorithm → Function (Mapping)
- Dispersive Flies Optimisation → Algorithm → Iteration
- Dispersive Flies Optimisation → Optimization Problem → Optimization
Neighborhood in Abstraction Space¶
Dispersive Flies Optimisation sits in a sparse region of the domain-specific corpus (94th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2551 abstractions)
Nearest neighbors
- Stochastic Tunneling — 0.79
- Kinetic Exchange Models of Markets — 0.78
- Evolutionary Algorithm — 0.78
- Searching the conformational space for docking — 0.78
- Marine Protected Area Network — 0.78
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
Particle Swarm Optimization has velocity and personal-best memory in the original paper's comparator; the printed DFO move has neither. The 2016 mammogram method is an adaptation with a different differential, and its results do not establish clinical diagnostic accuracy. A 2017 SVM paper repeats the baseline-looking move but prints a minimization selection rule while saying it maximizes F-measure, with no conversion explained; it is not a fully mapped second example of the exact baseline.[ref-89d9a4e9f710][ref-ee2c81317899][^ref-a178eb1cdc8e]
References¶
[^ref-89d9a4e9f710]: Mohammad Majid al-Rifaie, Dispersive Flies Optimisation, Proceedings of the Federated Conference on Computer Science and Information Systems (2014), pp.529–538, DOI 10.15439/2014F142; original full paper, §III Eq.(3) and Algorithm 1, printed pp.530–531; §§IV–V comparator and bounded experiments, pp.531–537. [^ref-1b4bd6e2f478]: Bishwa Babu Acharya, Sandeep Dhakal, Aayush Bhattarai and Nawraj Bhattarai, PID Speed Control of DC Motor Using Meta-Heuristic Algorithms, International Journal of Power Electronics and Drive Systems 12(2), 822–831 (2021), DOI 10.11591/ijpeds.v12.i2.pp822-831; original article PDF, abstract p.822; §2.4.2 pseudocode p.825; §2.5 Eq.(11), §2.6 Table 2 pp.826–827; §3.3 Tables 4–5 pp.827–829. Author-reported simulation, no implementation code inspected. [^ref-ee2c81317899]: Mohammad Majid al-Rifaie and Ahmed Aber, Dispersive Flies Optimisation and Medical Imaging, author manuscript for Recent Advances in Computational Optimization, Studies in Computational Intelligence 610, 183–203, DOI 10.1007/978-3-319-21133-6_11; §II Eq.(3) and §V.A adapted Eq.(9), PDF pp.1–2 and 8–9. This is an adaptation, not a baseline-rule positive. [^ref-a178eb1cdc8e]: Haya Abdullah Alhakbani and Mohammad Majid al-Rifaie, Optimising SVM to Classify Imbalanced Data Using Dispersive Flies Optimisation, Proceedings of the Federated Conference on Computer Science and Information Systems (2017), pp.399–402, DOI 10.15439/2017F91; original full paper, §II.A Algorithm 1 and §III, printed pp.400–401. The arg-min/F-measure-maximization inconsistency is unresolved.