Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Use of Parallel Explanatory Models to Enhance Transparency of Neural Network Configurations for Cell Degradation Detection

David Mulvey, Chuan Heng Foh, Muhammad Ali Imran, Rahim Tafazolli

TL;DR

The paper tackles the opacity of deep RNNs in detecting cell degradations within complex 5G networks. It introduces a parallel linearised model that operates in the pdf domain, assuming inputs can be represented as a Gaussian Mixture ($GMM$), to illuminate how spatial averaging and temporal processing shape distributions and affect fault detection. The key finding is that while main distributions improve with added layers, the emergence of sidelobes introduces new errors, explaining diminishing returns and guiding design toward simpler, efficient configurations. The approach also extends to higher-order and multi-layer RNNs and offers a pathway to broader application under the DARPA XAI framework, with potential extensions to LSTMs/GRUs and other sequential NN domains.

Abstract

In a previous paper, we have shown that a recurrent neural network (RNN) can be used to detect cellular network radio signal degradations accurately. We unexpectedly found, though, that accuracy gains diminished as we added layers to the RNN. To investigate this, in this paper, we build a parallel model to illuminate and understand the internal operation of neural networks, such as the RNN, which store their internal state in order to process sequential inputs. This model is widely applicable in that it can be used with any input domain where the inputs can be represented by a Gaussian mixture. By looking at the RNN processing from a probability density function perspective, we are able to show how each layer of the RNN transforms the input distributions to increase detection accuracy. At the same time we also discover a side effect acting to limit the improvement in accuracy. To demonstrate the fidelity of the model we validate it against each stage of RNN processing as well as the output predictions. As a result, we have been able to explain the reasons for the RNN performance limits with useful insights for future designs for RNNs and similar types of neural network.

Use of Parallel Explanatory Models to Enhance Transparency of Neural Network Configurations for Cell Degradation Detection

TL;DR

The paper tackles the opacity of deep RNNs in detecting cell degradations within complex 5G networks. It introduces a parallel linearised model that operates in the pdf domain, assuming inputs can be represented as a Gaussian Mixture (), to illuminate how spatial averaging and temporal processing shape distributions and affect fault detection. The key finding is that while main distributions improve with added layers, the emergence of sidelobes introduces new errors, explaining diminishing returns and guiding design toward simpler, efficient configurations. The approach also extends to higher-order and multi-layer RNNs and offers a pathway to broader application under the DARPA XAI framework, with potential extensions to LSTMs/GRUs and other sequential NN domains.

Abstract

In a previous paper, we have shown that a recurrent neural network (RNN) can be used to detect cellular network radio signal degradations accurately. We unexpectedly found, though, that accuracy gains diminished as we added layers to the RNN. To investigate this, in this paper, we build a parallel model to illuminate and understand the internal operation of neural networks, such as the RNN, which store their internal state in order to process sequential inputs. This model is widely applicable in that it can be used with any input domain where the inputs can be represented by a Gaussian mixture. By looking at the RNN processing from a probability density function perspective, we are able to show how each layer of the RNN transforms the input distributions to increase detection accuracy. At the same time we also discover a side effect acting to limit the improvement in accuracy. To demonstrate the fidelity of the model we validate it against each stage of RNN processing as well as the output predictions. As a result, we have been able to explain the reasons for the RNN performance limits with useful insights for future designs for RNNs and similar types of neural network.
Paper Structure (21 sections, 33 equations, 10 figures, 2 tables)

This paper contains 21 sections, 33 equations, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Modelling Approach
  4. Input Data
  5. Process Building Blocks
  6. Input Distributions
  7. Spatial Averaging
  8. Temporal Processing
  9. Multiple Layers
  10. Higher Order RNNs
  11. System Design and Implementation
  12. Network Simulation
  13. RNN Fault Detector
  14. Modelling Implementation
  15. Main Linearised Model
  16. ...and 6 more sections

Figures (10)

  • Figure 1: RNN and SVM Results for fault impact of 20dB.
  • Figure 2: System Diagrams of RNN and Models Showing Parallel Process Flow.
  • Figure 3: Illustration of RNN Inputs (15dB impact).
  • Figure 4: Illustration of $D_0$ per channel (15dB impact).
  • Figure 5: Relationship between Layer 2 and Layer 1 processing
  • ...and 5 more figures