Szegedy and colleagues: adversarial examples in neural networks

What changed

A surprising, reproducible class of classifier failures became an explicit evaluation target.

Scientific scope

Seven-author 2013 preprint revised in 2014 and conference publication; not a universal robustness theorem or all later adversarial ML.

Evidence

The linked sources identify this contribution and its scope. WRS has not independently reconstructed the scientific claims. The record is the eight-author 2014 conference paper. It establishes a failure phenomenon in the tested models and datasets, not a universal robustness theorem or the entirety of adversarial machine learning.

Contributors

• Christian Szegedy · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Wojciech Zaremba · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Ilya Sutskever · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Joan Bruna · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Dumitru Erhan · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Ian Goodfellow · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.
• Rob Fergus · SCIENCE: Named source author of the scoped work; individual effort shares are not measured.

• Szegedy et al., Intriguing properties of neural networks
• Google Research: Intriguing properties of neural networks

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction