Semantic Adversarial Deep Learning

Sanjit A. Seshia, Somesh Jha, and Tommaso Dreossi. Semantic Adversarial Deep Learning. IEEE Design and Test, 37(2):8–18, 2020.
Earlier version published in 2018 at CAV 2018.

Download

[HTML] 

Abstract

Models produced by machine-learning (ML) algorithms, especially deep neural networks, are being used in diverse domains where trustworthiness is a concern. The field of adversarial machine learning investigates the generation of inputs, termed adversarial examples, that cause the ML model to produce incorrect output. However, most existing approaches to generating adversarial examples and devising robust ML algorithms mostly ignore the semantics and context of the overall system containing the ML component. In addition, one may want to prioritize the search for adversarial examples towards those that significantly modify the desired semantics of the overall system. Along the same lines, existing algorithms for constructing robust ML algorithms ignore the specification of the overall system. Such considerations are particularly relevant in resource-constrained environments. In this paper, we present the paradigm of semantic adversarial machine learning, in which the semantics and specification of the overall system has a crucial role to play in this line of research. We present preliminary research results, and discuss directions for future work.

BibTeX

@article{seshia-ieeedt20,
  author    = {Sanjit A. Seshia and
               Somesh Jha and
               Tommaso Dreossi},
  title     = {Semantic Adversarial Deep Learning},
  journal   = {{IEEE} Design and Test},
  volume    = {37},
  number    = {2},
  pages     = {8--18},
  year      = {2020}
  abstract  = {Models produced by machine-learning (ML) algorithms, especially deep neural networks, are being used in diverse domains where trustworthiness is a concern. The field of adversarial machine learning investigates the generation of inputs, termed adversarial examples, that cause the ML model to produce incorrect output. However, most existing approaches to generating adversarial examples and devising robust ML algorithms mostly ignore the semantics and context of the overall system containing the ML component. In addition, one may want to prioritize the search for adversarial examples towards those that significantly modify the desired semantics of the overall system. Along the same lines, existing algorithms for constructing robust ML algorithms ignore the specification of the overall system. Such considerations are particularly relevant in resource-constrained environments. In this paper, we present the paradigm of semantic adversarial machine learning, in which the semantics and specification of the overall system has a crucial role to play in this line of research. We present preliminary research results, and discuss directions for future work.},
  wwwnote = {Earlier version published in 2018 at <a href="https://people.eecs.berkeley.edu/~sseshia/pubs/b2hd-dreossi-cav18.html">CAV 2018</a>.}  
}

Generated by bib2html.pl (written by Patrick Riley ) on Sun Aug 16, 2020 23:06:15