| name | metabolic-multi-agent-optimizer |
| description | Metabolic Multi-Agent Optimizer (MMAO) - bio-inspired optimization with endogenous resource allocation. Each agent carries internal energy with private-public metabolic loop. Fitness improvements converted to metabolic gains regulating sensing, search amplitude, branching, pruning, respawning. Parameter-light, self-calibrating. Use when: optimization without manual hyperparameter tuning, bio-inspired meta-heuristics, adaptive resource allocation in multi-agent systems. |
Core Methodology
Metabolic Resource Loop
Each agent maintains:
- Internal energy (private metabolic state)
- Continuous role state (search behavior identity)
- Motion or structural memory
- Local search history
Population shares:
- Communal resource pool (public metabolic state)
Closed-Loop Control
Fitness improvement → Normalized metabolic gain → Energy pool → Regulates:
├── Sensing intensity
├── Search amplitude
├── Role drift
├── Branching (replication)
├── Pruning (death)
├── Respawning (birth)
└── Elite reinvestment
Continuous Setting
- Energy-regulated symmetric zero-order probing
- Role-interpolated motion between agent types
Discrete Setting
- Structural sensing
- Local route improvement
- Guided perturbation
- Energy-weighted edge reuse
Key Properties
- Endogenous: Adaptation derived from internal metabolic loop, not externally attached modules
- Parameter-light: Self-calibrating through metabolic feedback
- Cross-domain: Same control law works for continuous and discrete optimization
- Validated: CEC2017 (10D/30D, 20 seeds) + TSPLIB (5 instances, 100 runs)
Progress Scale
Fitness improvements converted via:
- Robust progress scale (handles noisy improvements)
- Recent success statistic (short-term performance memory)
Implementation Pattern
class MetabolicAgent:
energy: float
role: float
memory: dict
def probe(self):
amplitude = f(self.energy)
return symmetric_zero_order_probe(amplitude)
def update_energy(self, fitness_delta):
gain = robust_progress_scale(fitness_delta)
self.energy += gain - metabolic_cost
When NOT to Use
- Not universally superior - main value is parameter-light self-calibration
- Use when manual hyperparameter tuning is costly, not when absolute performance is paramount
Activation
metabolic optimizer, bio-inspired optimization, multi-agent optimization, endogenous adaptation, parameter-light optimizer, MMAO, self-calibrating search, cs.NE