Deep Neural Networks Are Easily Fooled¶
Nguyen, A., Yosinski, J., & Clune, J. (2015). Deep Neural Networks Are Easily Fooled: High Confidence Predictions for Unrecognizable Images. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 427-436.
Cited by¶
2 citations across 2 artifacts.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Extrapolation Beyond Sampled Regime
- The softmax that says "panda, 99.7%" on noise is confidently wrong because the indicator tracks the regime, not the input's distance from it; catch it by asking whether the confidence was calibrated against inputs like this one.
This sourceShows trained classifiers emit near-certain softmax confidence (≥99.6%) on noise and unrecognizable inputs, demonstrating confidence is self-blind to regime exit.
Supported in partVerified against the work's full text
Nguyen et al. show DNNs assign near-certain softmax confidence to unrecognizable, noise-like images, but not the 'indicator tracks the regime' explanation, the 'panda, 99.7%' case, or a calibration check.
“believe to be recognizable objects with 99.99% confidence (e.g. labeling with certainty that white noise static is a lion)”
- The softmax that says "panda, 99.7%" on noise is confidently wrong because the indicator tracks the regime, not the input's distance from it; catch it by asking whether the confidence was calibrated against inputs like this one.
- Out Of Distribution Detection
- A closed-set classifier always answers; OOD detection adds the option to abstain. Not a confidence threshold alone. Low softmax confidence is not OOD detection: networks are notoriously confidently wrong on novel inputs.
This sourceDemonstrates DNNs assigning ≥99% softmax confidence to evolved, human-unrecognizable images — direct evidence that networks are confidently wrong on novel/garbage inputs.
Supported in partVerified against the work's full text
Shows DNNs giving near-certain softmax confidence to unrecognizable images, i.e. confidently wrong on synthetic inputs; says nothing about OOD detection, abstention or thresholds.
“We have demonstrated that discriminative DNN models are easily fooled in that they classify many unrecognizable images with near-certainty as members of a recognizable class.”
- A closed-set classifier always answers; OOD detection adds the option to abstain. Not a confidence threshold alone. Low softmax confidence is not OOD detection: networks are notoriously confidently wrong on novel inputs.
Verification¶
Does it exist? Confirmed. This work's DOI resolves to a registered record, which fixes its identity. That is all it fixes.
Does it back the claim? Read against the text for 2 of 2 citations: 2 supported in part. Each verdict is shown under its citation below, with what in the work backs the sentence.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Links previously used in the corpus¶
Before the registry existed this work was also linked 1 other way.
Registry ID ref:ba17df07d885 · see in the full table