| name | cortical-microcircuit-information-flux |
| description | Simulation-based reverse engineering methodology for analyzing whether cortical microcircuits are structurally organized to optimize information flux. Covers information flux quantification via mutual information, Recurrence Resonance mechanisms, core-embedding network architecture analysis, and bias-fluctuation contributions to neural dynamics. Applicable to: (1) biological neural circuit functional interpretation, (2) artificial recurrent system design including reservoir computers, (3) cortical microcolumn modeling, (4) information-theoretic analysis of neural networks. Activation: information flux, cortical microcircuit, reverse engineering neural networks, recurrence resonance, mutual information neural dynamics, core-embedding networks, reservoir computing optimization, brain information processing. |
Cortical Microcircuit Information Flux Optimization
Simulation-based reverse engineering study of whether cortical layer 5 microcircuits are structurally organized to enhance information flux in recurrent neural networks.
Core Concept
Information flux -- quantified by mutual information between successive network states -- is a prerequisite for rich information processing in recurrent neural networks. This methodology investigates whether biological cortical microcircuits are structurally optimized for this purpose.
Key Architecture Model
Layer 5 Cortical Microcolumn Model
[Core Population] <-- densely, strongly interconnected
|
[Embedding Network] <-- larger supporting network surrounding core
- Core neurons: Subset with high internal connectivity and strong coupling
- Embedding neurons: Remaining network providing contextual input
- Embedding effect: Pronounced flux-enhancing influence on core dynamics
Two Key Contributions of Embedding Network
1. Effective Biases
- Embedding network shifts core neurons into higher-entropy operating regime
- Biases move neurons away from saturated firing states
- Increases dynamic range available for information processing
- Can be quantified by comparing core activity with/without embedding
2. Stochastic Fluctuations via Recurrence Resonance
- Prevents core network from trapping in simple attractors (fixed points, oscillations)
- Recurrence Resonance: optimal noise level enhances signal transmission in recurrent systems
- Embedding provides structured noise that maintains rich dynamics
- Without embedding: core falls into low-complexity dynamical regimes
Analysis Methodology
Step 1: Information Flux Quantification
from scipy.stats import entropy
import numpy as np
def mutual_information_states(states_t, states_t1, bins=50):
"""Compute mutual information between successive network states."""
joint_hist, _, _ = np.histogram2d(
states_t.flatten(), states_t1.flatten(), bins=bins
)
p_xy = joint_hist / joint_hist.sum()
p_x = p_xy.sum(axis=1)
p_y = p_xy.sum(axis=0)
mi = 0
for i in range(len(p_x)):
for j in range(len(p_y)):
if p_xy[i, j] > 0:
mi += p_xy[i, j] * np.log(p_xy[i, j] / (p_x[i] * p_y[j]))
return mi
def information_flux(time_series, window=100):
"""Compute information flux over sliding windows."""
flux = []
for t in range(len(time_series) - window):
window_t = time_series[t:t+window]
window_t1 = time_series[t+1:t+window+1]
flux.append(mutual_information_states(window_t, window_t1))
return np.array(flux)
Step 2: Core-Embedding Decomposition
def decompose_core_embedding(activity_matrix, core_indices):
"""Separate core and embedding network activity."""
core_activity = activity_matrix[:, core_indices]
mask = ~np.isin(np.arange(activity_matrix.shape[1]), core_indices)
embedding_activity = activity_matrix[:, mask]
return core_activity, embedding_activity
def embedding_flux_contribution(full_flux, core_only_flux):
"""Quantify embedding network contribution to information flux."""
return full_flux - core_only_flux
Step 3: Reverse Engineering Analysis
def reverse_engineer_bias(activity_data, baseline_activity):
"""Extract effective biases exerted by embedding on core."""
mean_with_embedding = activity_data.mean(axis=0)
bias = mean_with_embedding - baseline_activity
return bias
def test_recurrence_resonance(noise_levels, core_activity):
"""Test if embedding noise operates at recurrence resonance optimum."""
flux_values = []
for noise_level in noise_levels:
perturbed = core_activity + noise_level * np.random.randn(*core_activity.shape)
flux = information_flux(perturbed)
flux_values.append(flux.mean())
optimal_idx = np.argmax(flux_values)
return noise_levels[optimal_idx], flux_values[optimal_idx]
Step 4: Self-Organization Principle
Biases can emerge from simple self-organization:
def self_organizing_bias_update(activity, target_entropy, learning_rate=0.01):
"""Update biases to achieve target entropy regime."""
current_entropy = entropy(activity)
error = target_entropy - current_entropy
bias_update = learning_rate * error * activity
return bias_update
Key Findings
- Embedding enhances flux: Core information flux significantly higher when embedded vs isolated
- Bias mechanism: Embedding shifts core into higher-entropy operating regime
- Fluctuation mechanism: Recurrence Resonance prevents attractor trapping
- Beyond biology: Individually optimized biases can increase flux beyond biological embedding
- Self-organization: Optimal biases can emerge from simple learning principles
Applications
Biological Interpretation
- Understanding why cortical circuits have their specific connectivity patterns
- Explaining the functional role of background neural activity
- Interpreting neuromodulation as bias/fluctuation control
Artificial System Design
- Reservoir computing: design embedding structures for optimal information processing
- RNN architectures: structured noise injection for maintaining rich dynamics
- Neuromorphic computing: bio-inspired connectivity patterns
Activation Keywords
- cortical microcircuit
- information flux
- recurrence resonance
- mutual information neural
- reverse engineering neural network
- reservoir computing optimization
- core-embedding architecture
- neural entropy optimization
