From Black-Box to White-Box: Control-Theoretic Neural Network Interpretability

TL;DR

This paper introduces a control-theoretic framework to interpret neural networks by modeling them as nonlinear state space systems, enabling analysis of neuron importance and internal dynamics for improved interpretability.

Contribution

It develops a novel method that applies control theory concepts to neural networks, providing a principled way to analyze neuron roles and internal modes.

Findings

Controllability measures neuron excitation ease.

Observability assesses neuron influence on output.

Hankel singular values rank internal modes by energy.

Abstract

Deep neural networks achieve state of the art performance but remain difficult to interpret mechanistically. In this work, we propose a control theoretic framework that treats a trained neural network as a nonlinear state space system and uses local linearization, controllability and observability Gramians, and Hankel singular values to analyze its internal computation. For a given input, we linearize the network around the corresponding hidden activation pattern and construct a state space model whose state consists of hidden neuron activations. The input state and state output Jacobians define local controllability and observability Gramians, from which we compute Hankel singular values and associated modes. These quantities provide a principled notion of neuron and pathway importance: controllability measures how easily each neuron can be excited by input perturbations, observability measures how strongly each neuron influences the output, and Hankel singular values rank internal modes that carry input output energy. We illustrate the framework on simple feedforward networks, including a 1 2 2 1 SwiGLU network and a 2 3 3 2 GELU network. By comparing different operating points, we show how activation saturation reduces controllability, shrinks the dominant Hankel singular value, and shifts the dominant internal mode to a different subset of neurons. The proposed method turns a neural network into a collection of local white box dynamical models and suggests which internal directions are natural candidates for pruning or constraints to improve interpretability.