Part IV · Ch. 25 — Swarm Intelligence
Evolutionary · Distributed optimization · Simple agents + local rules. Intelligent behavior emerges from many simple agents interacting locally. Card:
../02-types-of-ai/04-evolutionary.kmd.
🎨 Figure
F-IV.25.0— The whole greater than its parts. Brief: swarm of particles/ants following simple local rules, forming a global pattern (optimal path, formation); stylized pheromone trails.
1. Definition and short history
Optimization inspired by natural collectives (ants, bees, fish schools). It arose in the 1990s (PSO, ACO) for routing and logistics.
2. Foundations
- Biology / ethology — collective behavior, stigmergy.
- Complex systems — emergence, self-organization.
- Optimization — distributed population-based search.
- Probability — local stochastic rules.
3. Algorithms and architectures
- PSO (particle swarm) — particles follow their own best and the global best.
- ACO (ant colony) — pheromone reinforces good routes (stigmergy).
- Boids — separationalignmentcohesion rules (flocking simulation).
- Swarm robotics — many simple robots cooperating.
4. Inputs
- Hardware: CPU; naturally distributed/parallel.
- Data: objective function + problem topology.
- Data structures: agent population, pheromone map.
- Systems: swarm simulators; multi-robot platforms.
5. Specialized life cycle
| Stage | Specialization |
|---|---|
| 0 Problem | Distributed optimization/routing; multi-robot coordination |
| 1 Data | Define objective and local rules |
| 2 EDA | Structure of the search space |
| 3 Modeling | Choose PSO/ACO; interaction parameters |
| 4 "Training" | Iterate the swarm until convergence |
| 5 Evaluation | Solution quality, speed, robustness |
| 5.5 Acceptance | Stability, tolerance to agent failures |
| 6 Production | Run the swarm (optimization or robots) |
| 7 Monitoring | Convergence, agent failures |
| 8 Retraining | Readjust parameters/rules |
| 9 Governance | Safety of multi-robot systems |
6. Capabilities, modes, and modalities
Collective/distributed: routing (telecom, logistics), optimization, multi-robot coordination, crowd simulation; robust to failures (no single point).
7. Limits, risks, and ethics
Convergence not guaranteed; parameter tuning sensitive; emergent behavior hard to predict/control; safety in physical swarms.
8. State of the art and examples
ACOPSO in optimization and routing, boids in simulationanimation, swarm robotics and coordinated drones; a niche of distributed, fault-tolerant problems.
Evolutionary Paradigm complete (ch. 23–25). With it, Part IV closes the 34 chapters. Index:
INDEX.kmd.