Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multiscale Astrocyte Network Calcium Dynamics for Biologically Plausible Intelligence in Anomaly Detection

Berk Iskar, Michael Taynnan Barros

TL;DR

This work tackles the challenge of adapting anomaly detectors to evolving threats by introducing a Ca$^{2+}$-modulated learning framework that couples a mesoscopic astrocyte Ca$^{2+}$ field to a feedforward DNN. The Ca$^{2+}$ field models IP$_3$-mediated CICR, SERCA uptake, and conductance-aware gap-junction diffusion on a 3D astrocyte lattice, with the resulting signals mapped to synapses via a mass-preserving operator and smoothed to form learning cues. A Ca$^{2+}$-gated update rule, with a slowly adapting Ca$^{2+}$ threshold (metaplasticity) and a Laplacian-based heterosynaptic regularizer, gates plasticity during training, while precomputing Ca trajectories keeps inference overhead negligible. Empirically, the Ca$^{2+}$-gated DNN (Ca-DNN) outperforms a matched baseline on the CTU-13 (Neris) dataset across multiple train/test splits, achieving up to approximately 99% accuracy and reducing false positives/negatives, demonstrating the utility of biologically grounded, context-sensitive plasticity for streaming anomaly detection.

Abstract

Network anomaly detection systems encounter several challenges with traditional detectors trained offline. They become susceptible to concept drift and new threats such as zero-day or polymorphic attacks. To address this limitation, we propose a Ca$^{2+}$-modulated learning framework that draws inspiration from astrocytic Ca$^{2+}$ signaling in the brain, where rapid, context-sensitive adaptation enables robust information processing. Our approach couples a multicellular astrocyte dynamics simulator with a deep neural network (DNN). The simulator models astrocytic Ca$^{2+}$ dynamics through three key mechanisms: IP$_3$-mediated Ca$^{2+}$ release, SERCA pump uptake, and conductance-aware diffusion through gap junctions between cells. Evaluation of our proposed network on CTU-13 (Neris) network traffic data demonstrates the effectiveness of our biologically plausible approach. The Ca$^{2+}$-gated model outperforms a matched baseline DNN, achieving up to $\sim$98\% accuracy with reduced false positives and negatives across multiple train/test splits. Importantly, this improved performance comes with negligible runtime overhead once Ca$^{2+}$ trajectories are precomputed. While demonstrated here for cybersecurity applications, this Ca$^{2+}$-modulated learning framework offers a generic solution for streaming detection tasks that require rapid, biologically grounded adaptation to evolving data patterns.

Multiscale Astrocyte Network Calcium Dynamics for Biologically Plausible Intelligence in Anomaly Detection

TL;DR

This work tackles the challenge of adapting anomaly detectors to evolving threats by introducing a Ca-modulated learning framework that couples a mesoscopic astrocyte Ca field to a feedforward DNN. The Ca field models IP-mediated CICR, SERCA uptake, and conductance-aware gap-junction diffusion on a 3D astrocyte lattice, with the resulting signals mapped to synapses via a mass-preserving operator and smoothed to form learning cues. A Ca-gated update rule, with a slowly adapting Ca threshold (metaplasticity) and a Laplacian-based heterosynaptic regularizer, gates plasticity during training, while precomputing Ca trajectories keeps inference overhead negligible. Empirically, the Ca-gated DNN (Ca-DNN) outperforms a matched baseline on the CTU-13 (Neris) dataset across multiple train/test splits, achieving up to approximately 99% accuracy and reducing false positives/negatives, demonstrating the utility of biologically grounded, context-sensitive plasticity for streaming anomaly detection.

Abstract

Network anomaly detection systems encounter several challenges with traditional detectors trained offline. They become susceptible to concept drift and new threats such as zero-day or polymorphic attacks. To address this limitation, we propose a Ca-modulated learning framework that draws inspiration from astrocytic Ca signaling in the brain, where rapid, context-sensitive adaptation enables robust information processing. Our approach couples a multicellular astrocyte dynamics simulator with a deep neural network (DNN). The simulator models astrocytic Ca dynamics through three key mechanisms: IP-mediated Ca release, SERCA pump uptake, and conductance-aware diffusion through gap junctions between cells. Evaluation of our proposed network on CTU-13 (Neris) network traffic data demonstrates the effectiveness of our biologically plausible approach. The Ca-gated model outperforms a matched baseline DNN, achieving up to 98\% accuracy with reduced false positives and negatives across multiple train/test splits. Importantly, this improved performance comes with negligible runtime overhead once Ca trajectories are precomputed. While demonstrated here for cybersecurity applications, this Ca-modulated learning framework offers a generic solution for streaming detection tasks that require rapid, biologically grounded adaptation to evolving data patterns.

Paper Structure

This paper contains 16 sections, 29 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Model and Methodology
  3. Framework Overview
  4. Multicellular Ca$^{2+}$ Field (Mesoscopic Model)
  5. Gap-Junction State Dynamics and Conductance
  6. From Ca$^{2+}$ Field to Learning Signals
  7. Ca$^{2+}$-Weighted Plasticity With Adaptive Thresholds
  8. Ca$^{2+}$-Gated Weight Dynamics on a DNN
  9. Mutual Information Evaluation
  10. Results
  11. Multicellular Ca$^{2+}$ Field Across Runs
  12. Transmitter--Receiver Mutual Information
  13. Sensitivity and Learning Performance
  14. End-to-End Detection on CTU-13 (Neris)
  15. Discussion
  16. ...and 1 more sections

Figures (5)

  • Figure 1: Aggregated $\mathrm{Ca}^{2+}$ per cell (1--54) for 7 runs; each curve = one run. Panels (a,c) include exogenous forced spikes (10 cells/run), (b,d) do not. Panels (c,d) extend transmitter duration.
  • Figure 2: (a) Spatial decay of $I_\star$ vs. graph distance (mean $\pm$95% CI). (b) Predictive relevance: higher $I_\star$$\Rightarrow$ larger accuracy gains.
  • Figure 3: (a) Parameter settings and accuracies across runs. (b) Pearson correlation matrix between accuracy and Ca$^{2+}$ parameters. (c) Standardized regression coefficients indicating each parameter’s relative impact.
  • Figure 4: Accuracy across data splits (mean of 10 runs). (a) 8k/8k, (b) 10k/10k, (c) 7k/3k (70/30), (d) 6k/4k (60/40). Ca-DNN = Ca$^{2+}$-gated; Base = baseline DNN.
  • Figure 5: Confusion matrices after (a) 10 epochs and (b) 30 epochs. Darker cells indicate higher counts; the color scale is shown on the right.