Get the latest Science News and Discoveries

Scientists uncover quantum-inspired vulnerabilities in neural networks: the role of conjugate variables in system attacks - EurekAlert


<p>In a recent study published in National Science Review, researchers from the Northwest Institute of Nuclear Technology and Xi&#39;an Jiaotong University have utilized the mathematical principles from quantum mechanics to analyze the vulnerabilities in neural networks. The study reveals that neural networks are inherently susceptible to small, non-random perturbations, similar to the mathematics in quantum uncertainty principle. This similarity shows that attackers can exploit these vulnerabilities by manipulating conjugate variables related to the network&#39;s inputs.</p>

None

Get the Android app

Or read this on Eureka Alert

Read more on:

Photo of Neural Networks

Neural Networks

Photo of system attacks

system attacks

Photo of conjugate variables

conjugate variables

Related news:

News photo

HKU astrophysicists discover a novel method for hunting the first stars - EurekAlert

News photo

Giving local business a 'boost' with national lab technology - EurekAlert

News photo

Unprecedented deep oxidative desulfurization with precisely designed Ti sites - EurekAlert