VU SoS Lablet Quarterly Executive Summary - January 2020

A. Fundamental Research

The Science of Security for Cyber-Physical Systems (CPS) Lablet focuses on (1) Foundations of CPS Resilience, (2) Analytics for CPS Cybersecurity, (3) Development of a Multi-model Testbed for Simulation–based Evaluation of Resilience, and (4) Mixed Initiative and Collaborative Learning in Adversarial Environments. 

• We assess the vulnerabilities of load forecast systems for smart grid based on neural networks and propose a defense mechanism to construct resilient forecasters. We model the strategic interaction between a defender and an attacker as a Stackelberg game, where the defender decides first the prediction scheme and the attacker chooses afterwards its attack strategy. Here, the defender selects randomly the sensor measurements to use in the forecast, while the adversary calculates a bias to inject in some sensors.  We find an approximate equilibrium of the game and implement the defense mechanism using an ensemble of predictors, which introduces uncertainties that mitigate the attack's impact. We evaluate our defense approach training forecasters using data from an electric distribution system simulated in GridLAB-D.
• At the Year 2, Quarter 2 meeting we presented the methods work so far and illustrated the processes and results with reference to the network view of the Tallinn Manual 2.0 on the International Law Applicable to Cyber Operations, a quasi-legal document of nearly 600 pages. Two questions were raised: (1) Can this method be applied to the NIST Privacy Framework, and, (2) Can we provide an assessment of if the current work can be applied to the challenge of understanding the structure and processes articulated in the text of the NIST Privacy Framework. The team concluded that current research work and methodologies developed can be implemented in the draft Privacy Framework, subject to availability of specific data pertaining to the sub-categories. Further, in the event that the required data are not available, we identified alternative approaches that can help bridge the gap between data-in-hand and data-requested/required.
• We are working on a Jupyter Notebook integration capability in WebGME, which would allow Python-based data exploration and/or model modifications to be implemented. The current notebook-based workflow and the key architectural elements are as follows: A (Javascript-based) WebGME plugin can programmatically generate Jupyter Notebooks, based on the contents of the model. Note, that this plugin can translate the visual (WebGME) model to an arbitrary Python data structure (e.g. NetworkX graph). Alternatively, the Python code inside the Notebook can access the WebGME server via its REST API to query elements of the model. The Jupyter Notebook server (co-hosted with the WebGME server) is accessed with a simple iFrame-based visualizer inside the WebGME interface. With generated notebook code developers can implement custom analysis algorithms and may send data back to WebGME (modify the model). The model modification is supported by another (Javascript-based) plugin that has direct access to the model
• In many settings where multiple agents interact, the optimal choices for each agent depend heavily on the choices of the others. These coupled interactions are well-described by a general-sum differential game, in which players have differing objectives, the state evolves in continuous time, and optimal play is characterized by Nash equilibria. Often, problems admit multiple Nash equilibria.  From the perspective of a single agent in such a game, this multiplicity of solutions can introduce uncertainty about how other agents will behave. This paper proposes a general framework for resolving ambiguity between Nash equilibria by reasoning about the equilibrium other agents are aiming for. We demonstrate this framework in simulations of a multi-player human-robot navigation problem that yields two main conclusions: First, by inferring which equilibrium humans are operating at, the robot is able to predict trajectories more accurately, and second, by discovering and aligning itself to this equilibrium the robot is able to reduce the cost for all players.

B. Community Engagement(s)

• Our research was presented in the 2019 Conference on Decision and Game Theory for Security (GameSec 2019) and also in Resilience Week 2019.

C. Educational Advances

• We are developing a new course in systems theory at UC Berkeley, to be taken by upper level undergraduates and first and second year graduate students, on a rapprochement between control theory and reinforcement learning.  The course will focus on a modern viewpoint on modeling, analysis, and control design, leveraging tools and successes from both systems and control theory and machine learning.  The first version of this course will be taught by Shankar Sastry in Spring 2020.