Visible to the public 

Machine learning (ML) is proposed as a solution to scalable defensive and offensive capabilities in cyber security. The proposals range from semi-automated decision support tools to fully-automated capabilities. However, ML models can be exploited in at least four ways:

  1. attackers can poison training data used to train ML algorithms to degrade prediction quality, or redirect predictions, altogether;
  2. attackers can evade by manipulating runtime data to ensure ML models misclassify malicious behavior as benign;
  3. attackers can infer records in the training data; and
  4. attackers can approximately reconstruct ML models.

In the MAML project we are developing a framework consisting of:

  1. a lightweight simulation language to express the performance parameters and architecture tailored to represent a decision-support environment consisting of one or more ML models and decision support tool users;
  2. metrics to measure the quality of an adversarial influence strategy conducted in a simulation; and
  3. mitigations, including ML model design guidelines to improve resiliency against attacks.

As participants of the MAML kickoff we are working towards the goal of Coordination, Gap Analysis, and FY19 & Out Plans. Please respond to the following survey to be discussed immediately upon commencement of the kickoff the morning of 09/20/2018:

We very much appreciate your time and effort in helping us get MAML off to a great start!

Dan Clouse - MAML PI