Explainable Neural Computation via Stack Neural Module Networks
This work addresses the need for interpretable reasoning processes in complex inferential tasks like question answering, offering a novel method that reduces reliance on supervised traces.
The paper tackles the challenge of enabling compositional reasoning in machine learning models while ensuring interpretability, by introducing a neural modular approach that automatically induces sub-task decompositions without strong supervision, resulting in improved interpretability for human evaluators.
In complex inferential tasks like question answering, machine learning models must confront two challenges: the need to implement a compositional reasoning process, and, in many applications, the need for this reasoning process to be interpretable to assist users in both development and prediction. Existing models designed to produce interpretable traces of their decision-making process typically require these traces to be supervised at training time. In this paper, we present a novel neural modular approach that performs compositional reasoning by automatically inducing a desired sub-task decomposition without relying on strong supervision. Our model allows linking different reasoning tasks though shared modules that handle common routines across tasks. Experiments show that the model is more interpretable to human evaluators compared to other state-of-the-art models: users can better understand the model's underlying reasoning procedure and predict when it will succeed or fail based on observing its intermediate outputs.