On the Computational Entanglement of Distant Features in Adversarial Machine Learning
This work addresses adversarial robustness in neural networks, offering a novel perspective on feature behavior, though it appears incremental in applying entanglement to a specific adversarial context.
The paper tackled the problem of adversarial machine learning by introducing 'computational entanglement' in overparameterized linear networks, which enables fitting random noise to achieve zero loss and transforms worst-case adversarial examples into recognizable, robust outputs, revealing insights into non-robust features.
In this research, we introduce the concept of "computational entanglement," a phenomenon observed in overparameterized feedforward linear networks that enables the network to achieve zero loss by fitting random noise, even on previously unseen test samples. Analyzing this behavior through spacetime diagrams reveals its connection to length contraction, where both training and test samples converge toward a shared normalized point within a flat Riemannian manifold. Moreover, we present a novel application of computational entanglement in transforming a worst-case adversarial examples-inputs that are highly non-robust and uninterpretable to human observers-into outputs that are both recognizable and robust. This provides new insights into the behavior of non-robust features in adversarial example generation, underscoring the critical role of computational entanglement in enhancing model robustness and advancing our understanding of neural networks in adversarial contexts.