QUINN (Quantum INspired Neural Networks)

Machine learning models are susceptible to adversarial attacks that make modifications to the input data in order to cause misclassifications. The root cause is the linearity of the decision boundaries of machine learning models in relation to their inputs. One promising direction is to represent the input data as a distribution. Quantum information science entails techniques for working with wave functions, which are related to probability distributions over a given space. Representing the input data as a wave function could lead to improved defenses against adversarial attacks. Thus, SoarTech proposes Quantum INspired Neural Networks (QUINN), a quantum machine learning framework robust to adversarial attacks. SoarTech envisions QUINN as a wrapper around an arbitrary neural network that protects against adversarial attacks by detecting modified data indicative of an attack, identifying the parts of the data that were modified, and cleansing the data of any modifications. This is accomplished by converting incoming data into a wave function representation and using the concepts of entanglement and decoherence to compare incoming data with the distribution of legitimate data in various ways. The end result will be a system hardened against adversarial attacks that can be deployed in a wide range of domains.