NICOP - Provenance Analytics for Command and Control (PA4C2)
Abstract
Provenance, traditionally used to establish the quality of food and authenticity of art, isdemonstrating similar strong benefits for data by describing flows of data in systems andproviding the basis for accountability and trust. However, as systems log provenanceassertions, and as processes are running 24/7, users are very quickly confronted with aprovenance data deluge, making it challenging to grasp what is happening in those systems,whether there are events they need investigating, and whether performance is adequate.To this end, we have developed original analytics techniques over provenance data, helpingextract knowledge out of them and enabling users to gain insights into a system s behaviorand the data it generates.In this project, we propose to apply and extend such analytics techniques to work onprovenance captured in command and control (C2) systems for autonomous vehicles. Thework program will address key research questions about the potential benefits ofprovenance in the planning and task identification in the IMPACT autonomous vehiclemanagement system developed Space and Naval Warfare Systems (SPAWAR) Center Pacific.Capturing the provenance of plans and actions taken in IMPACT brings a number of benefits.First, such provenance information creates the foundation for the accountability ofautonomous systems by providing explicit documentation of actions taken by them and,crucially, the inputs on which such actions depend. This information not only providesexplanations for an unmanned vehicle~s behavior while disconnected but also allows thecentral command to detect inconsistencies across systems and identify where reconciliationis required. Secondly, provenance pattern analysis may highlight abnormal behaviors ofvehicles, which may indicate a faulty condition or, especially after an extended period ofdisconnection, a high risk of being tampered with. Thirdly, given that the quality of certaindata can be derived from its provenance information, it may also be possible to infer thequality of auto-generated plans from their provenance. Such a quality indication for planswill be a useful aid to human operators deciding which plan to be assigned to vehiclesIn brief, the objectives of the project are:1. To design provenance analytics techniques that combine domain-agnostic properties ofprovenance, such as network metrics and topology, with C2 domain-specificinformation, to relate provenance logs about the creation of plans to quality of plans interms of their outcomes.2. To devise analytics techniques capable of mining provenance and extracting discreteactivities performed by operators and unmanned vehicle, and compare the performanceof these techniques.3. To devise principles of accountability allowing the system and the analytics techniquesto be accountable to their stakeholders in the context of unmanned vehiclemanagement systems.4. To study the scalability of the techniques investigated in the project.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 10, 2018
- Source ID
- N629091812079
Entities
People
- Luc Moreau
Organizations
- Office of Naval Research
- United States Navy