**Lillian J. Ratliff**

**ratliffl at eecs.berkeley **

Societal-scale infrastructures spanning energy, healthcare, and transportation are at an important inflection point in their operations due to increased interdependence on new cyber-physical system (CPS) technologies such as wireless sensor/actuator networks, data-driven real-time learning techniques being implemented in the cloud, and ubiquitous mobile computing devices for intermediating between networks of wireless sensors and the cloud. The advent of these societal-scale cyber-physical systems (S-CPS) brings with it new opportunities for improving efficiency while simultaneously exposing novel vulnerabilities.

My research interests lie at the intersection of non-cooperative games, incentive design, statistical learning, and societal scale cyber-physical systems (S-CPS). In particular, I am interested in designing economic incentives and control in order to balance the efficiency-vulnerability tradeoff inherent to S-CPS.

Owing to the scale of wireless sensor-actuator networks, existence of machine-to-machine interfaces operating in real-time, and large numbers of decision makers competing over oft–scarce resources, game theory and data-driven analytics do not alone suffice to capture the fundamental properties of intelligent infrastructures—game theory generally assumes knowledge of the environment and data-driven analytics cannot capture competitive interactions. However, both are necessary; game theory captures complex socioeconomic interactions while statistical learning aids in reducing dimensionality and provides tools for analysis of large-scale, time-series data. I am interested in developing data-driven models for system and agent behavior in S-CPS and utilizing these models to design incentives for behavior modification. Further, pervasive sensing and mis-aligned incentives give rise to new vulnerabilities such as privacy and security issues. In my research I aim to strike a balance between these vulnerabilities and the efficiency gained by utilizing the underlying CPS infrastructure in S-CPS. Currently I am working on designing privacy-preserving and resilient incentive strategies for sustainability in S-CPS.

The aim of this project is to utilize tools from differential topology and
geometry to study the interaction of competitive agents. In particular, by
looking at games in which the strategy spaces are non-convex (either finite or
infinite dimensional), we may develop
useful decompositions in order to study existence of equilibria and their
properties in this more abstract setting. Further, we have developed a
characterization of local Nash equilibria—**differential Nash
equilibria**—that is generic, structurally stable,
and amenable to computation thereby allowing for the replacement of
computationally intractable inequalities defining Nash equilibria in both the
incentive design and utility learning problems with conditions that only need to
be checked at a single point. The goal is to use the
analytical and computational advancements to developing novel schemes for
decentralized control in engineered systems as well as onâ€“line identification
techniques for human agents. We are currently working on methods to decompose
games into the *non-cooperative* and *cooperative* components thereby
enabling us to find sufficient descent directions for more efficient
computation of differential Nash. (Collaborator: Samuel Burden [UC
Berkeley])

We model the interaction between non-cooperative agents and a social planner as a reversed Stackelberg game where the social planner is the leader and the agents are the followers that play simultaneously. The social planner would like to modify the agents' behavior so that it aligns with some desired behavior which allows the social planner to meet her objective. However, we assume that the social planner does not know the agents' utility functions and hence must learn them in order to design incentives which modify the agents behavior. Using the differential Nash concept, we are able to cast the problem of solving for the parameters of the agents' utility functions and incentive mechanism as convex optimization problems making the method scalable. We enforce a second-order, non-degeneracy condition resulting in the leader-induced differential Nash equilibrium being isolated—hence, robust to measurement noise and modeling errors.

Through a **social game** we integrate building occupants into the
control and management of an office building that is instrumented
with networked embedded systems for sensing and actuation. The goal
of the social game is to both incentivize building occupants to be
more energy efficient and learn behavioral models for occupants so
that the building can be made *resilient* and *sustainable* through automation. The
social game for energy savings that we have designed is such that
occupants in an office building vote according to their usage
preferences of shared resources and are rewarded with points based
on how energy efficient their strategy is in comparison with the
other occupants. Having points increases the likelihood of the
occupant winning in a lottery. We have employed tools from
statistical learning theory to develop an algorithm for computing
the stopping time given a specified level of accuracy, an underlying
game-theoretic generative model for behavior, and a estimation or
classification task, e.g. estimating the distribution of
players' utility function parameters,
classifying players into categories, forecasting energy usage—useful for
demand response. (Collaborators: Ming Jin [UC Berkeley], Ioannis
Konstantakopoulos [UC Berkeley], Costas Spanos [UC Berkeley])

The goal of this project is to design pricing mechanisms to coordinate non-cooperative agents. Applications that are considered include efficient energy management in buildings, controlled diffusion in multi-agent networks, and network security. In a non-cooperative game theoretic framework, the goal is to design pricing schemes to close the gap between the decentralized and the centralized cost. In addition, we are interested in designing robust pricing mechanisms. In an effort to do so, we first characterize when feedback Nash strategies are stable and use this characterization to inform the design of pricing mechanisms. (Collaborators: Sam Coogan [UC Berkeley], Daniel Calderone [UC Berkeley])

Disaggregation is the process of taking an aggregated signal and
decomposing it into its components. The application we are
interested in is energy disaggregation where the aggregate signal is
the power consumption for a whole building (residential or
commercial) and the goal is to decompose the aggregate signal into
the power signal for all the contributing devices. By providing the
disaggregated data to consumers, we are enabling them to construct
methods for energy efficient operation of the building.
Disaggregation of consumer energy data evokes **privacy** concerns.
We study the inherent trade-off between privacy and the utility of
the data.
(Collaborators: Aaron
Bestick [UC Berkeley], Henrik
Ohlsson [C3 Energy, UC
Berkeley], Roy
Dong [UC Berkeley])

Privacy is ontologically subjective in that what privacy means is interpreted by the individual. However, this does not mean we cannot make quantitative statements about privacy. We do this by first constructing inference metrics which quantify the ability of an adversary to make inferences and then by utilizing economic tools that allow us to understand individual preferences over privacy viewed as an economic good. This combination of statistical inference/detection theory along with game theory provides a natural framework for understanding privacy in S-CPS. Specifically in the context of energy S-CPS, by examining the fundamental limits of non-intrusive load monitoring, we developed a privacy metric that quantifies the ability of adversaries to make inferences about the consumer. We use this privacy metric for designing privacy-based service contracts in which privacy is viewed as a good and electricity service is offered as a product line differentiated according to privacy where consumers can self-select the level of privacy that fits their needs and wallet. We study the effects of risk and the distribution of types—the former has impact on the security-insurance tradeoff and the latter on privacy settings offered—on social welfare and efficiency. (Collaborators: Henrik Ohlsson[C3 Energy, UC Berkeley], Roy Dong, Carlos Barreto [UT Dallas], Alvaro A. Cárdenas [UT Dallas])

We are currently revisiting the microscopic to macroscopic view of multi-lane traffic with a game theoretic view of lane-changing by taking a physics-based, ground-up approach and employing tools from the theory of mean-field games to study the limit as the number of vehicles grows large. This game-theoretic model can be leveraged in the design of tolling and incentive schemes based on driver or lane type. (Collaborator: Dan Calderone [UC Berkeley])

CDC, LA, Dec. 2014, Organizing and speaking at workshop on 'Big Data Analytics for Societal Scale Cyber-Physical Systems: Energy Systems'

Allerton, IL, Oct. 2014, Social Game for Building Energy Efficiency: Incentive Design.

IFAC, Cape Town, South Africa, August 2014, Incentive Design and Utility Learning via Energy Disaggregation.

ACC, Portland Oregon, 06 June 2014, Genericty and Structural Stability of Non-degenerate Differential Nash Equilibria

Stanford, 30 May 2014, Game Theoretic Tools for Societal Scale Cyber-Physical Systems

PDE Seminar, Math Department Berkeley, 09 May 2014, Convexity in Multidimensional Screening

PDE Seminar, Math Department Berkeley, 25 April 2014, Multidimensional Screening

EECS Seminar, University of Michigan, 25 March 2014, Game Theoretic Tools for Societal Scale CPS

** Lillian J. Ratliff**, Roy Dong, Walid Krichene, and S. Shankar Sastry. "Utility Learning and Incentive Design
via Adaptive Control," 2015.

** Lillian J. Ratliff**, Ming Jin, Ioannis Konstantakopoulos,
Costas Spanos, and S. Shankar Sastry. "Utility Learning and Incentive Design
for Sustainability in Building Energy Management," 2015.

**Lillian J. Ratliff**, Samuel A. Burden, S. Shankar Sastry. "On
the Characterization of Local Nash Equilibria in Continuous Games."
IEEE Transactions on Automatic Control, 2014.*
(under review)* (arXiv:1411.2168)

** Lillian J. Ratliff**, Carlos Barreto, Roy Dong, Henrik
Ohlsson, Alvaro A. Cárdenas, and S. Shankar Sastry. Effects of
Risk on Privacy Contracts for Demand-Side Management. IEEE
Transactions on Smart Grid, 2014. * Under Review, 2nd Round (
Apr. 2015)* (arXiv:1409.7926v3)

Pushkin Kachroo, **Lillian J. Ratliff**, and S. Shankar Sastry.
"Analysis of the Godunov Based Hybrid Model for Ramp Metering and
Robust Feedback Control Design." IEEE Transactions on Intelligent
Transportation Systems, Vol. 15, Issue 5, 2132-2142, Oct. 2014. (
PDF)

Roy Dong, Alvaro A. Cárdenas, ** Lillian J. Ratliff**, Henrik
Ohlsson, Shankar Sastry. "Quantifying the Utility-Privacy Tradeoff in
the Smart Grid." IEEE Transactions on Smart Grid, 2014. *Under
Review, 2nd Round
(Apr. 2015).* (
arXiv:1406.2568v1)

Pushkin Kachroo, **Lillian J. Ratliff**, and S. Shankar Sastry.
"A New Static Traffic Assignment Using Density Based Travel Time."
Applied Mathematical Modeling, 2014. * Under Review (July). *

Dexter Scobee, **Lillian Ratliff**, Roy Dong, Henrik Ohlsson,
Michel Verhaegen and S. Shankar Sastry. "Nuclear Norm Minimization for
Blind Subspace Identification (N2BSID)," CDC 2015.
(*Submitted*)

Daniel Calderone, **Lillian Ratliff**, S. Shankar Sastry. "Lane
Pricing via Decisionâ€“Theoretic Lane Changing Model of Driver
Behavior," CDC 2015. (*Submitted*)

Ming Jin, ** Lillian J. Ratliff**, Ioannis C. Konstantakopoulos,
Costas Spanos, S. Shankar
Sastry. "REST: A Reliable Estimation and Stopping Time Algorithm for
Social Game Experiments."
ACM/IEEE International Conference on Cyber-Physical Systems (ICCPS),
2015. ( PDF)

** Lillian J. Ratliff**, Ming Jin, Ioannis C. Konstantakopoulos,
Costas Spanos, S. Shankar
Sastry. "Social Game for Building Energy Efficiency: Incentive Design."
52nd Annual Allerton Conference on Communication, Control, and Computing, 2014. ( PDF, Slides)

** Lillian J. Ratliff**, Roy Dong, Henrik Ohlsson, Alvaro A.
Cardenas, Shankar Sastry. "Privacy and Customer Segmentation in the
Smart Grid." IEEE Conference on Decision and Control, 2014. ( PDF)

** Lillian J. Ratliff**, Roy Dong, Henrik Ohlsson, S. Shankar
Sastry. "Incentive Design and Utility Learning via Energy
Disaggregation." 19th World Congress of the International Federation
of Automatic Control (IFAC), 2014. (PDF)

Daniel J. Calderone, ** Lillian J. Ratliff**, Shankar Sastry.
"Pricing for Coordination in Open-Loop Differential Games." 19th World Congress of the International Federation of Automatic Control (IFAC),
2014. (PDF)

Henrik Ohlsson, ** Lillian J. Ratliff**, Roy Dong, S. Shankar
Sastry. "Blind Identification via Lifting." 19th World Congress of the
International Federation of Automatic Control (IFAC), 2014. (PDF)

Roy Dong, ** Lillian J. Ratliff**, Henrik Ohlsson, Shankar
Sastry. "Fundamental Limits of Non-Intrusive Load Monitoring."
ACM International Conference on High Confidence Networked Systems (HiCoNS), 2014. DOI:
10.1145/2566468.2566471 (arXiv)

** Lillian J. Ratliff**, Samuel A. Burden, Shankar Sastry.
"Genericity and Structural Stability of Non-degenerate Differential
Nash Equilibria." American Control Conference, 2014. DOI: 10.1109/ACC.2014.6858848
(PDF)

** Lillian J. Ratliff**, Samuel A. Burden, Shankar Sastry.
"Characterization and Computation of Local Nash Equilibria in
Continuous Games." 51st Annual Allerton Conference on Communication,
Control, and Computing, 2013. DOI: 10.1109/Allerton.2013.6736623 (PDF, Slides)

Roy Dong, ** Lillian J. Ratliff**, Henrik Ohlsson, S. Shankar
Sastry. "Energy Disaggregation via Adaptive Filtering." 51st Annual
Allerton Conference on Communication, Control, and Computing, 2013.
DOI: 10.1109/Allerton.2013.6736521 (arXiv)

Roy Dong, **Lillian J. Ratliff**, Henrik Ohlsson, Shankar
Sastry. "A Dynamical Systems Approach to Energy Disaggregation."
IEEE Conference on Decision and Control, 2013. pg. 6335-6340. DOI: 10.1109/CDC.2013.6760891
(arXiv, PDF)

Daniel Calderone, **Lillian J. Ratliff**, S. Shankar Sastry.
"Pricing Design for Robustness in Linear-Quadratic Dynamic Games."
IEEE Conference on Decision and Control, 2013. DOI: 10.1109/CDC.2013.6760558

Aaron Bestick, **Lillian J. Ratliff**, Po Yan, Ruzena Bajcsy,
S. Shankar Sastry. "An Inverse Correlated Equilibrium Framework for
Utility Learning in Multiplayer, Noncooperative Settings." ACM
International Conference on High Confidence Networked Systems
(HiCoNS), 2013. DOI: 10.1145/2461446.2461449 (PDF)

Samuel Coogan, **Lillian J. Ratliff**, Daniel Calderone,
Claire Tomlin, S. Shankar Sastry. "Energy Management via Pricing in
LQ Dynamic Games." American Control Conference, 2013. DOI: 10.1109/ACC.2013.6579877 (PDF)

**Lillian J. Ratliff**, Samuel Coogan, Daniel Calderone, S. Shankar Sastry. " Pricing in Linear-Quadratic Dynamic Games." 50th Annual Allerton Conference on Communication, Control, and Computing, 2012. DOI: 10.1109/Allerton.2012.6483440 (PDF)

**Lillian J. Ratliff** and Pushkin Kachroo. "Validating
numerically consistent macroscopic traffic models using microscopic
data." Transportation Research Board 89th Annual Meeting, 2010.

Daniel P. Cook, Yitung Chen, **Lillian J. Ratliff**, Huajun Chen, and Jian
Ma. "Numerical Modeling of EM Pump Efficiency." ASME Conference Proceedings, 2006.

Henrik Ohlsson,** Lillian Ratliff**, Roy Dong, Shankar Sastry. "Blind Identification of ARX Models with Piecewise Constant Inputs." * arXiv:1303.6719, 2013.* (PDF)

** Lillian J. Ratliff**, Roy Dong, Henrik Ohlsson, and S. Shankar Sastry.
Energy Efficiency via Incentive Design and Utility Learning. ACM International
Conference on High Confidence Networked Systems (HiCoNS), 2014. DOI:
10.1145/2566468.2576849
(PDF)

**Lillian J. Ratliff**, Samuel A. Burden, S. Shankar Sastry. "Characterization and Computation of Local Nash Equilibria in Continuous Games." at TRUST Annual Conference, Washington, D.C. April, 2013. (PDF)

**Lillian Ratliff**, Daniel Calderone, Samuel Coogan, S. Shankar Sastry. "Pricing for Coordination in Dynamic Games." at FORCES Kickoff Meeting, Washington, D.C. April, 2013. (PDF)

Daniel Calderone, Samuel Coogan, **Lillian Ratliff**, Anil Aswani, Claire
Tomlin, S. Shankar Sastry. "Quadratic incentive design for noncooperative
distributed control" * ACM International Conference on High Confidence Networked Systems (HiCoNS)* at CPS Week 2012, Beijing, China, May 2012. (PDF)

NSF Graduate Fellowship, 2009

MS thesis topic: Uncertainty Propagation in Dynamical Systems.

Advisor: Pushkin Kachroo