Diagram Index

Visual guides to understanding TWEANN concepts and architecture.

Core Concepts

TWEANN Structure

The fundamental architecture of a Topology and Weight Evolving Artificial Neural Network.

Key elements:

• Sensors (green): Input layer receiving environmental signals
• Hidden neurons (blue): Processing nodes with evolving topology
• Actuators (orange): Output layer producing actions
• Connections: Weighted links that can be added/removed through evolution
• Cortex: Coordinator process managing sync cycles

Neuroevolution Cycle

The iterative process of population-based optimization.

The four phases:

1. Population: Collection of neural network genotypes
2. Evaluation: Run each phenotype, calculate fitness scores
3. Selection: Tournament or truncation selection of survivors
4. Reproduction: Mutation, crossover, and elitism to create next generation

NEAT Evolution

NeuroEvolution of Augmenting Topologies - how structure evolves.

Key innovations:

• add_node: Split a connection to insert a new neuron
• add_link: Create a new connection between existing nodes
• Innovation numbers: Historical markings for meaningful crossover
• Speciation: Group similar topologies to protect innovation

Architecture

Genotype to Phenotype

Transformation from genetic encoding to living neural network.

The Constructor pattern:

• Genotype (left): Records stored in ETS (agent, cortex, sensor, actuator, neuron)
• Phenotype (right): Concurrent Erlang processes communicating via messages
• Each neuron becomes a gen_server with its own state and plasticity

Supervision Tree

OTP supervision hierarchy for fault tolerance.

Module Dependencies

How the library modules relate to each other.

C4 Architecture Model

Software architecture using the C4 model.

Context Diagram

Container Diagram

Component Diagram

LTC Neurons

LTC Neuron Architecture

Liquid Time-Constant neurons with temporal dynamics.

Components:

• Time constant (tau) controls adaptation speed
• Closed-form approximation (CfC) for efficient computation
• Temporal memory through leaky integration

LTC vs Standard Neurons

Comparison between LTC and traditional neurons.

Learning Mechanisms

Neural Plasticity

Online weight learning during the network's lifetime.

Plasticity rules:

• Hebbian: "Cells that fire together, wire together"
• Oja's Rule: Self-normalizing Hebbian (performs online PCA)
• Self-Modulation: Neuron controls its own learning parameters
• Neuromodulation: External reward signal gates learning

Activation Functions

The transfer functions available for neurons.

Function families:

• Bounded: sigmoid, tanh (classification, gates)
• Unbounded: relu, softplus (deep networks, sparse activation)
• Periodic: sin, cos (rhythmic patterns, CPGs)
• Localized: gaussian (RBF networks, pattern recognition)
• Binary: step, sgn (threshold logic)

Mutation Sequence

Step-by-step mutation process.

Evaluation Cycle Sequence

Detailed evaluation workflow.

Distributed Evolution

Distributed Evolution Model

Multi-node evolution architecture.

Federated Populations Model

Island-based distributed populations.

Swarm Evolution Model

Swarm intelligence with evolved controllers.

Mega-Brain Architecture

Large-scale distributed neural architectures.

Application Domains

Military Swarm Coordination

Autonomous swarm systems.

Civil Infrastructure Resilience

Infrastructure protection applications.

Counter-Drone Detection Fusion

Multi-sensor fusion for detection.

Layered Defense Zones

Multi-layer defense architecture.

See Also

• TWEANN Basics - Introduction to neuroevolution concepts
• Architecture Details - In-depth system architecture
• LTC Neurons - Liquid Time-Constant neuron guide
• Quick Start - Get started with examples