Intriguing properties of neural networks
E108000
"Intriguing properties of neural networks" is a highly influential research paper that revealed surprising vulnerabilities and behaviors of deep neural networks, particularly their susceptibility to adversarial examples.
All labels observed (1)
| Label | Occurrences |
|---|---|
| Intriguing properties of neural networks canonical | 1 |
How this entity was disambiguated
This entity first appeared as the object of triple T921634 — resolving that mention is where its identity was fixed. The disambiguator weighed these candidate entities and picked the highlighted one (or “None”, minting a new entity). This is how homonymy is resolved: the same surface form can point to different entities.
NED1
Entity disambiguation (via context triple)
gpt-5-mini-2025-08-07
Target entity: Intriguing properties of neural networks Context triple: [Christian Szegedy, notableWork, Intriguing properties of neural networks]
-
A.
Hopfield networks
Hopfield networks are recurrent artificial neural networks that serve as content-addressable memory systems, storing patterns as stable states and retrieving them through dynamics that minimize an energy function.
-
B.
“Learning representations by back-propagating errors”
“Learning representations by back-propagating errors” is a landmark 1986 research paper that popularized the backpropagation algorithm for training multi-layer neural networks, helping to launch the modern field of deep learning.
-
C.
Boltzmann machines
Boltzmann machines are stochastic recurrent neural networks used for learning complex probability distributions, foundational in unsupervised learning and energy-based models.
-
D.
Deep belief networks
Deep belief networks are a class of deep generative neural network models composed of stacked layers of latent variables, typically built from restricted Boltzmann machines, used for unsupervised feature learning and representation.
-
E.
“A fast learning algorithm for deep belief nets”
“A fast learning algorithm for deep belief nets” is a seminal 2006 paper by Geoffrey Hinton that introduced an efficient unsupervised pretraining method for deep neural networks using stacked restricted Boltzmann machines.
- F. None of above. chosen
- G. Unsure - the case is ambiguous/there is not enough information to decide.
NED2
Entity disambiguation (via description)
gpt-5-mini-2025-08-07
Target entity: Intriguing properties of neural networks Target entity description: "Intriguing properties of neural networks" is a highly influential research paper that revealed surprising vulnerabilities and behaviors of deep neural networks, particularly their susceptibility to adversarial examples.
-
A.
Hopfield networks
Hopfield networks are recurrent artificial neural networks that serve as content-addressable memory systems, storing patterns as stable states and retrieving them through dynamics that minimize an energy function.
-
B.
“Learning representations by back-propagating errors”
“Learning representations by back-propagating errors” is a landmark 1986 research paper that popularized the backpropagation algorithm for training multi-layer neural networks, helping to launch the modern field of deep learning.
-
C.
Boltzmann machines
Boltzmann machines are stochastic recurrent neural networks used for learning complex probability distributions, foundational in unsupervised learning and energy-based models.
-
D.
Deep belief networks
Deep belief networks are a class of deep generative neural network models composed of stacked layers of latent variables, typically built from restricted Boltzmann machines, used for unsupervised feature learning and representation.
-
E.
“A fast learning algorithm for deep belief nets”
“A fast learning algorithm for deep belief nets” is a seminal 2006 paper by Geoffrey Hinton that introduced an efficient unsupervised pretraining method for deep neural networks using stacked restricted Boltzmann machines.
- F. None of above. chosen
Statements (46)
| Predicate | Object |
|---|---|
| instanceOf |
research paper
ⓘ
scientific article ⓘ |
| argues | neural networks are too linear in high-dimensional spaces ⓘ |
| arxivId | arXiv:1312.6199 ⓘ |
| author |
Christian Szegedy
ⓘ
Dumitru Erhan ⓘ Ian Goodfellow ⓘ Ilya Sutskever ⓘ Joan Bruna ⓘ Rob Fergus ⓘ Wojciech Zaremba ⓘ |
| citationContext | often cited as the first major work on adversarial examples in deep learning ⓘ |
| concludes | adversarial examples are a fundamental property of neural networks ⓘ |
| considered | highly influential in deep learning research ⓘ |
| demonstratesOn | ImageNet-like image classification tasks ⓘ |
| field |
computer vision
ⓘ
deep learning ⓘ machine learning ⓘ |
| focusesOn |
adversarial examples
ⓘ
deep neural networks ⓘ image classification ⓘ model robustness ⓘ neural networks ⓘ |
| hasTopic |
high-dimensional geometry of neural networks
ⓘ
stability of deep learning models ⓘ transferability of adversarial examples ⓘ vulnerability of neural networks ⓘ |
| influenced |
development of adversarial training methods
ⓘ
research on adversarial machine learning ⓘ research on robustness of deep learning models ⓘ security analysis of machine learning systems ⓘ |
| institution |
Google
ⓘ
New York University ⓘ Université de Montréal ⓘ |
| language | English ⓘ |
| publishedAs | arXiv preprint ⓘ |
| shows |
adversarial examples can transfer between different architectures
ⓘ
adversarial examples can transfer between models trained on different subsets of data ⓘ adversarial examples can transfer between models trained with different hyperparameters ⓘ adversarial examples generalize across different models ⓘ existence of adversarial examples for deep networks ⓘ linear behavior in high-dimensional spaces contributes to adversarial vulnerability ⓘ small imperceptible perturbations can cause misclassification ⓘ |
| uses |
convolutional neural networks
ⓘ
image recognition benchmarks ⓘ |
| year | 2013 ⓘ |
How these facts were elicited
The pipeline generated the facts above by prompting gpt-5.1 with this entity's name + description and the instruction below.
Instruction
You are a knowledge base construction expert. Given a subject entity and a description of it, return factual statements that you know for the subject as a JSON list of dictionaries(triples), where keys must be "subject", "predicate" and "object". The number of facts may be very high, between 25 to 50 or more, for very popular subjects. For less popular subjects, the number of facts can be very low, like 5 or 10. # Requirements - If you don't know the subject at all, return an empty list. - If the subject is not a named entity, return an empty list. - Include at least one triple where predicate is "instanceOf". - Do not get too wordy. - Separate several objects into multiple triples with one object.
Input
Subject: Intriguing properties of neural networks Description of subject: "Intriguing properties of neural networks" is a highly influential research paper that revealed surprising vulnerabilities and behaviors of deep neural networks, particularly their susceptibility to adversarial examples.
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.