Biologically Inspired Mechanisms for Adversarial Robustness
A convolutional neural network strongly robust to adversarial perturbations at reasonable computational and performance cost has not yet been demonstrated. The primate visual ventral stream seems to be robust to small perturbations in visual stimuli but the underlying mechanisms that give rise to th...
Main Authors: | , , , |
---|---|
Format: | Technical Report |
Published: |
Center for Brains, Minds and Machines (CBMM)
2020
|
Online Access: | https://hdl.handle.net/1721.1/125981 |
_version_ | 1811072658065850368 |
---|---|
author | Vuyyuru Reddy, Manish Banburski, Andrzej Plant, Nishka Poggio, Tomaso |
author_facet | Vuyyuru Reddy, Manish Banburski, Andrzej Plant, Nishka Poggio, Tomaso |
author_sort | Vuyyuru Reddy, Manish |
collection | MIT |
description | A convolutional neural network strongly robust to adversarial perturbations at reasonable computational and performance cost has not yet been demonstrated. The primate visual ventral stream seems to be robust to small perturbations in visual stimuli but the underlying mechanisms that give rise to this robust perception are not understood. In this work, we investigate the role of two biologically plausible mechanisms in adversarial robustness. We demonstrate that the non-uniform sampling performed by the primate retina and the presence of multiple receptive fields with a range of receptive field sizes at each eccentricity improve the robustness of neural networks to small adversarial perturbations. We verify that these two mechanisms do not suffer from gradient obfuscation and study their contribution to adversarial robustness through ablation studies. |
first_indexed | 2024-09-23T09:09:30Z |
format | Technical Report |
id | mit-1721.1/125981 |
institution | Massachusetts Institute of Technology |
last_indexed | 2024-09-23T09:09:30Z |
publishDate | 2020 |
publisher | Center for Brains, Minds and Machines (CBMM) |
record_format | dspace |
spelling | mit-1721.1/1259812020-07-31T10:36:47Z Biologically Inspired Mechanisms for Adversarial Robustness Vuyyuru Reddy, Manish Banburski, Andrzej Plant, Nishka Poggio, Tomaso A convolutional neural network strongly robust to adversarial perturbations at reasonable computational and performance cost has not yet been demonstrated. The primate visual ventral stream seems to be robust to small perturbations in visual stimuli but the underlying mechanisms that give rise to this robust perception are not understood. In this work, we investigate the role of two biologically plausible mechanisms in adversarial robustness. We demonstrate that the non-uniform sampling performed by the primate retina and the presence of multiple receptive fields with a range of receptive field sizes at each eccentricity improve the robustness of neural networks to small adversarial perturbations. We verify that these two mechanisms do not suffer from gradient obfuscation and study their contribution to adversarial robustness through ablation studies. This material is based upon work supported by the Center for Brains, Minds and Machines (CBMM), funded by NSF STC award CCF-1231216. 2020-06-25T15:39:29Z 2020-06-25T15:39:29Z 2020-06-23 Technical Report Working Paper Other https://hdl.handle.net/1721.1/125981 CBMM Memo;110 application/pdf Center for Brains, Minds and Machines (CBMM) |
spellingShingle | Vuyyuru Reddy, Manish Banburski, Andrzej Plant, Nishka Poggio, Tomaso Biologically Inspired Mechanisms for Adversarial Robustness |
title | Biologically Inspired Mechanisms for Adversarial Robustness |
title_full | Biologically Inspired Mechanisms for Adversarial Robustness |
title_fullStr | Biologically Inspired Mechanisms for Adversarial Robustness |
title_full_unstemmed | Biologically Inspired Mechanisms for Adversarial Robustness |
title_short | Biologically Inspired Mechanisms for Adversarial Robustness |
title_sort | biologically inspired mechanisms for adversarial robustness |
url | https://hdl.handle.net/1721.1/125981 |
work_keys_str_mv | AT vuyyurureddymanish biologicallyinspiredmechanismsforadversarialrobustness AT banburskiandrzej biologicallyinspiredmechanismsforadversarialrobustness AT plantnishka biologicallyinspiredmechanismsforadversarialrobustness AT poggiotomaso biologicallyinspiredmechanismsforadversarialrobustness |