Corpus ID: 604334

Intriguing properties of neural networks

  title={Intriguing properties of neural networks},
  author={Christian Szegedy and W. Zaremba and Ilya Sutskever and Joan Bruna and D. Erhan and Ian J. Goodfellow and R. Fergus},
  • Christian Szegedy, W. Zaremba, +4 authors R. Fergus
  • Published 2014
  • Computer Science
  • CoRR
  • Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. [...] Key Result In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.Expand Abstract

    Figures, Tables, and Topics from this paper.

    Intriguing Properties of Learned Representations
    • 2018
    Feature-Robustness, Flatness and Generalization Error for Deep Neural Networks
    Dense Associative Memory Is Robust to Adversarial Inputs
    • 49
    • Open Access
    Exploring LOTS in Deep Neural Networks
    Towards Distortion-Predictable Embedding of Neural Networks
    On the Blindspots of Convolutional Networks
    • 3
    • Highly Influenced
    • Open Access
    Deep neural networks are easily fooled: High confidence predictions for unrecognizable images
    • 1,658
    • Highly Influenced
    • Open Access
    How deep is deep enough? - Optimizing deep neural network architecture
    • 1
    • Open Access


    Publications referenced by this paper.
    Measuring Invariances in Deep Networks
    • 353
    • Open Access
    Learning Deep Architectures for AI
    • 6,136
    • Open Access
    Building high-level features using large scale unsupervised learning
    • 1,869
    • Highly Influential
    • Open Access
    Visualizing and Understanding Convolutional Networks
    • 8,252
    • Open Access
    ImageNet classification with deep convolutional neural networks
    • 51,776
    • Open Access
    ImageNet Classification with Deep Convolutional Neural Networks
    • 51,776
    • Open Access
    Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
    • 11,339
    • Open Access
    Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups
    • 5,883
    • Open Access