Detection Instability
The Detection Instability task evaluates whether an evolved Dynamic Neural Field (DNF) architecture can reliably detect a transient external stimulus and propagate this detection to an output field.
The desired behaviour is:
- An external stimulus induces a localised activation peak in the input field.
- This peak is self-stabilized during stimulus presentation.
- Activity is transmitted to the output field, which also forms a corresponding peak.
- After stimulus offset, both fields relax back to baseline (no persistent memory).
This task tests the most fundamental Dynamic Field Theory mechanism: stimulus-driven detection without memory.
The highest-fitness solutions consistently converge to a minimal topology:
- Fields: 2 (input → output)
- Hidden fields: 0
- Connections: 1 excitatory feedforward coupling
- Dynamics: Gaussian lateral interaction kernels within each field
The figure above shows:
- Blue: activity during stimulus presentation
- Red: activity after stimulus removal
Both fields exhibit a stimulus-driven peak that disappears once the input is removed, matching the detection instability criterion.
Analysed 100 runs; 100 reached the fitness threshold (100.0% success, threshold = 0.950)
- Mean: 1.00
- Median: 1.00
- Std: 0.00
- Mean convergence rate (fitness gain/gen): 0.2267
- Mean fitness improvement/gen: 0.2267
- Hidden fields (mean ± std): 0.00 ± 0.00
- Enabled connections (mean ± std): 1.00 ± 0.00
| label | run | value |
|---|---|---|
| Highest max fitness | 2026-01-23 10h11m26s | 0.974337 |
| Lowest max fitness | 2026-01-23 10h28m18s | 0.959235 |
| Fastest to threshold | 2026-01-23 09h46m06s | 1.0 |
| Slowest to threshold | 2026-01-23 09h46m06s | 1.0 |
| Most hidden fields | 2026-01-23 09h46m06s | 0.0 |
| Most enabled connections | 2026-01-23 09h46m06s | 1.0 |
This section reports detailed statistics from the evolutionary run that produced the highest-fitness solution for the Detection Instability task.
Final generation: g = 1
- Best fitness: 0.9743
- Target fitness: 0.950
- Average fitness: 0.7251
Overall run statistics
- Max best fitness: 0.9743 (reached at generation 1)
- Mean best fitness over run (AUC): 0.8561
- Mean average fitness over run (AUC): 0.6699
- Longest stagnation period: 0 generations
- Target fitness first reached: generation 1 (best fitness ≈ 0.9743)
Final generation (g = 1)
- Species: 2
- Active species: 2
Across run
- Total distinct species created: 1
- Species extinct by final generation: 0
- Average species per generation: 1.50
- Average active species per generation: 1.50
- Max active species in a generation: 2 (at g = 1)
Species lifetime & size
- Average species lifespan: 1.00 generations
- Longest-lived species: 0 (lifespan 1)
- Average max members per species: 1000.00
- Average offspring per species: 0.00
Final generation (g = 1)
- Avg genome size: 2.15
- Avg field genes: 2.00
- Avg connection genes: 0.15
Growth over run
- Genome size change: +0.15 (≈ +0.146 / gen)
- Field genes change: +0.00 (≈ +0.000 / gen)
- Connection genes change: +0.15 (≈ +0.146 / gen)
Ratios
- Avg connections per field (final gen): 0.07
-
Field kernels:
- Gaussian: 1609 (80.5%)
- Mexican-hat: 391 (19.6%)
-
Interaction kernels: none
Generation: 1 Fitness: 0.9743
| Field | Role | Kernel type | Field parameters | Kernel parameters |
|---|---|---|---|---|
| nf 1 | Input | Gaussian | h = -14.98, τ = 91.03 | A = 14.04, σ = 9.32, A_glob = -0.19 |
| nf 2 | Output | Gaussian | h = -2.12, τ = 84.91 | A = 5.95, σ = 9.82, A_glob = -0.13 |
| Interaction gene | From → To | Kernel parameters |
|---|---|---|
| gk cg 1-2-0 | nf 1 → nf 2 | A = 20.49, σ = 3.70, A_glob = -0.05 |
detection.mp4
The best Detection Instability solution:
- Emerged immediately (generation 1)
- Required no speciation pressure
- Exhibited minimal structural growth
- Used only Gaussian kernels for both fields and interactions
This confirms that detection behaviour lies at the base of the evolutionary search space, serving as a stable attractor for neat-dnfs.
Detection Instability is trivially solvable for neat-dnfs:
- Evolution converges in a single generation.
- The minimal architecture is sufficient and consistently rediscovered.
- This task serves as a baseline sanity check for stimulus propagation and transient peak formation.