Neural network interpretation using descrambler groups

TL;DR

This paper introduces a group-theoretical method to interpret deep neural networks, transforming their internal signals into human-readable features, demonstrated on a DSP network where it revealed complex internal structures.

Contribution

The paper presents a novel group-theoretical approach to descramble neural network signals, enhancing interpretability especially in scientific computing and DSP applications.

Findings

Successfully descrambled a DSP neural network, revealing internal structures.

Discovered the network's spontaneous invention of various signal processing features.

Achieved results in ten minutes of training from a random start.

Abstract

The lack of interpretability and trust is a much-criticised feature of deep neural networks. In fully connected nets, the signalling between inner layers is scrambled because backpropagation training does not require perceptrons to be arranged in any particular order. The result is a black box; this problem is particularly severe in scientific computing and digital signal processing (DSP), where neutral nets perform abstract mathematical transformations that do not reduce to features or concepts. We present here a group-theoretical procedure that attempts to bring inner layer signalling into a human-readable form, the assumption being that this form exists and has identifiable and quantifiable features - for example, smoothness or locality. We applied the proposed method to DEERNet (a DSP network used in electron spin resonance) and managed to descramble it. We found considerable internal sophistication: the network spontaneously invents a bandpass filter, a notch filter, a frequency axis rectifier, frequency division multiplexing, group embedding, spectral filtering regularisation, and a Fourier-like transform that maps harmonic functions into Chebyshev polynomials - in ten minutes of unattended training from a random initial guess.