MITEI 2019 Research Poster

Design a service-oriented architecture that permits private DERs to behave in accordance with the grid’s overall interests in addition to their own.

A Service-Oriented Architecture for the Electric Distribution Grid

MIT 6.867 2019 Poster

In broad terms, our project explores adversarial attacks on machine learning models. We do so first in a toy “gaussian” setting, then in an image CIFAR-10 setting.

  • Toy setting: a “adversary” attempts to uncover a distribution hidden by a “challenger”.
  • Image Setting: training data reconstruction. We aim to reconstruct a single datapoint (image) within the dataset that the target model was trained on.

Adversarial Attacks on Machine Learning Models