Verified70% confidenceFactExact time
After superimposing tiny perturbations that are nearly invisible to the human eye onto an image correctly classified by a neural network, the model output will produce an error with extremely high confidence.
4
Sources
70%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Anti-Surveillance Fashion: How Adversarial Examples Can Make Cameras Fail to Recognize You
hackernewshackernews7/5/2026
Related Claims
UnverifiedAdversarial sample attacks can add noise to images that is nearly imperceptible to the human eye yet causes models to produce incorrect classifications80% similarUnverifiedShifting an image by one pixel can cause a CNN model's recognition result to change drastically or even fail completely.77% similarUnverifiedIn neuroscience scenarios, electron microscope images require a combination of manual and automatic annotation, and annotation quality directly determines model performance; neuron segmentation models are prone to errors in regions with blurred boundaries; and model evaluation involves how to measure segmentation accuracy and how to detect systematic misjudgments.76% similarUnverifiedAI可见性仪表盘用有限的预设问题推断整体可见性，存在极大的采样偏差76% similarUnverifiedThe concept of adversarial examples was first systematically articulated by Ian Goodfellow et al. in 2014, demonstrating that adding imperceptibly small perturbations to images in the computer vision domain can cause classifiers to produce incorrect judgments.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/204509API
curl https://kongchang.com/api/v1/knowledge/claims/204509MCP
get_claim(id=204509)