Developmental Silo Guide

The Developmental Silo manages ontogeny - how individuals develop over their lifetime. Instead of networks being born fully formed, they grow through stages, experience critical periods, and can undergo metamorphosis. This mirrors biological development.

Overview

Traditional neuroevolution treats networks as static entities. The Developmental Silo introduces lifetime dynamics:

• Developmental stages - Progress from embryonic to adult to senescent
• Plasticity decay - Learning capacity decreases with age
• Critical periods - Time-limited windows for specific learning
• Metamorphosis - Radical structural reorganization
• Canalization - Traits become increasingly stable

Why Developmental Dynamics?

Architecture

The Developmental Silo uses TWEANN controllers at three levels:

Sensors (10 inputs)

The L0 controller receives 10 developmental measurements:

Actuators (8 outputs)

The controller adjusts 8 developmental parameters:

Developmental Dynamics

Developmental Stages

Individuals progress through five life stages:

Each stage has:

• Different mutation rate modifiers
• Available critical period types
• Appropriate plasticity ranges

Plasticity Decay

Plasticity decreases with age, following exponential decay:

plasticity(t+1) = plasticity(t) * (1 - decay_rate)

The min_plasticity floor (typically 0.1) prevents complete rigidity.

Critical Periods

Critical periods are time-limited windows where learning is enhanced:

During critical period:

effective_learning_rate = base_learning_rate * 2.0

Critical periods:

• Open automatically based on stage
• Close after critical_period_duration generations
• Cannot be reopened once closed
• Record what learning was acquired

Metamorphosis

Metamorphosis allows radical transformation:

%% Metamorphosis resets plasticity and reduces canalization
new_plasticity = initial_plasticity * severity + old_plasticity * (1 - severity)
new_canalization = old_canalization * (1 - severity)

Metamorphosis triggers:

• Maturation progress exceeds metamorphosis_trigger threshold
• Maximum 3 metamorphoses per lifetime (max_metamorphoses)
• 20% probability when conditions met

Canalization

Canalization measures resistance to perturbation:

Canalization increases with age and successful behavior, but metamorphosis resets it.

Heterochrony

Heterochrony measures developmental timing relative to population:

• Precocious (early developers) - Mature faster, reach adult stage sooner
• Delayed (late developers) - Extended juvenile/adolescent periods

The heterochrony_index sensor tracks this:

z_score = abs(individual_age - population_mean_age) / population_std_age
heterochrony_index = clamp(z_score / 4.0 + 0.5, 0.0, 1.0)

Integration with Other Silos

Outgoing Signals

Incoming Signals

Signal Examples

%% Receive stress from ecological silo
handle_cast({cross_silo, stress_level, Stress}, State) when Stress > 0.7 ->
    %% High stress accelerates maturation
    NewSpeed = State#developmental_state.maturation_speed * 1.5,
    {noreply, State#developmental_state{maturation_speed = min(0.1, NewSpeed)}};

%% Send plasticity signal to cultural silo
signal_plasticity_available(Pid) ->
    IndStates = maps:values(State#developmental_state.individual_states),
    Plasticities = [I#individual_dev_state.plasticity || I <- IndStates],
    AvgPlasticity = lists:sum(Plasticities) / length(Plasticities),
    clamp(AvgPlasticity, 0.0, 1.0).

Events Emitted

Practical Examples

Example 1: Fast-Maturing Population

Configure for rapid development and short critical periods:

Config = #{
    growth_rate => 0.15,                 % Fast structural growth
    maturation_speed => 0.08,            % Fast physiological maturation
    critical_period_duration => 3,        % Short learning windows
    plasticity_decay_rate => 0.12,        % Fast plasticity loss
    initial_plasticity => 0.95            % High starting plasticity
}.

Expected outcomes:

• Quick convergence to adult behavior
• Less lifetime adaptation
• Efficient for stable deployment environments

Example 2: Extended Learning

Configure for prolonged development and extended learning:

Config = #{
    growth_rate => 0.03,                 % Slow structural growth
    maturation_speed => 0.01,            % Slow maturation
    critical_period_duration => 15,       % Long learning windows
    plasticity_decay_rate => 0.02,        % Slow plasticity loss
    initial_plasticity => 0.9,
    metamorphosis_trigger => 0.9,         % High metamorphosis threshold
    metamorphosis_severity => 0.7         % Strong metamorphosis effect
}.

Expected outcomes:

• Extended learning periods
• High lifetime adaptation
• Better for changing deployment environments

Example 3: Deployment Adaptation

For networks that need to adapt in deployment:

Config = #{
    plasticity_decay_rate => 0.01,        % Very slow decay
    initial_plasticity => 0.8,
    metamorphosis_trigger => 0.5,         % Easy to trigger metamorphosis
    metamorphosis_severity => 0.4,        % Moderate reset
    developmental_noise_level => 0.15     % Some stochasticity
}.

Expected outcomes:

• Networks retain learning capacity in deployment
• Can adapt to environment-specific conditions
• Metamorphosis available for major adaptation

Tuning Guide

Trade-offs

Common Issues

Recommended Starting Point

DefaultConfig = #{
    growth_rate => 0.05,
    maturation_speed => 0.02,
    critical_period_duration => 5,
    plasticity_decay_rate => 0.05,
    initial_plasticity => 0.9,
    developmental_noise_level => 0.1,
    metamorphosis_trigger => 0.8,
    metamorphosis_severity => 0.5
}.

Control Loop

The Developmental Silo executes per generation:

1. Advance individual ages - Increment age counter
2. Apply plasticity decay - Reduce plasticity based on decay rate
3. Check stage transitions - Move to next stage if conditions met
4. Open/close critical periods - Manage time-limited learning windows
5. Check metamorphosis triggers - Initiate radical transformation if threshold met
6. Update canalization levels - Track trait stability
7. Emit developmental events - Notify other systems
8. Send cross-silo signals - Update dependent silos

Configuration Reference

Full Configuration Record

-record(developmental_config, {
    %% Enable/disable
    enabled = true :: boolean(),

    %% Stage configuration
    num_stages = 5 :: pos_integer(),
    stage_definitions :: [developmental_stage()],

    %% Plasticity
    min_plasticity = 0.1 :: float(),
    max_plasticity = 1.0 :: float(),

    %% Critical periods
    enable_critical_periods = true :: boolean(),
    available_period_types = [sensory, motor, cognitive, social] :: [atom()],

    %% Metamorphosis
    enable_metamorphosis = true :: boolean(),
    max_metamorphoses = 3 :: pos_integer(),

    %% Event emission
    emit_events = true :: boolean()
}).

API Functions

%% Start the developmental silo
developmental_silo:start_link(Config)

%% Register a new individual
developmental_silo:register_individual(Pid, IndividualId)

%% Advance an individual's age
{ok, NewAge} = developmental_silo:advance_age(Pid, IndividualId)

%% Get plasticity level
Plasticity = developmental_silo:get_plasticity(Pid, IndividualId)

%% Open a critical period
ok = developmental_silo:open_critical_period(Pid, IndividualId, sensory)

%% Close a critical period
ok = developmental_silo:close_critical_period(Pid, IndividualId, sensory)

%% Trigger metamorphosis
ok = developmental_silo:trigger_metamorphosis(Pid, IndividualId)

%% Get individual developmental state
IndState = developmental_silo:get_individual_state(Pid, IndividualId)

%% Enable/disable silo
developmental_silo:enable(Pid)
developmental_silo:disable(Pid)

Source Code Reference

Core implementation files:

References

• PLAN_DEVELOPMENTAL_SILO.md - Full specification
• "Developmental Plasticity and Evolution" - West-Eberhard
• "Critical Periods in Development" - Knudsen
• "Metamorphosis: Postembryonic Reprogramming" - Shi

See Also

• Cultural Silo Guide - Learning and imitation
• Ecological Silo Guide - Environmental stress effects
• Morphological Silo Guide - Network structure growth
• Temporal Silo Guide - Time management in development