UROP Openings

Have a UROP opening you would like to submit?

Please fill out the form.

Submit your UROP opening

Biologically Inspired Mechanisms for Adversarial Robustness


Term:

Fall

Department:

9: Brain and Cognitive Sciences

Faculty Supervisor:

Tomaso Poggio

Faculty email:

tp@ai.mit.edu

Apply by:

09/15/20

Contact:

Andy Banburski, kappa666@mit.edu

Project Description

A convolutional neural network strongly robust to adversarial perturbations at reasonable computational and performance cost has not yet been demonstrated. We know that 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 project, we will investigate the role of biologically plausible mechanisms in adversarial robustness. We have previously demonstrated 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 now want to move beyond these simple mechanisms and study the effect of attention for selecting fixations, as well as work on a predictive generative model that would act as a sort of "anomaly detector".

Pre-requisites

- Knowledge of Tensorflow or Pytorch - prior experience in Deep Learning research is a bonus