As AI systems are integrated into high stakes social domains, researchers now examine how to desi... more As AI systems are integrated into high stakes social domains, researchers now examine how to design and operate them in a safe and ethical manner. However, the criteria for identifying and diagnosing safety risks in complex social contexts remain unclear and contested. In this paper, we examine the vagueness in debates about the safety and ethical behavior of AI systems. We show how this vagueness cannot be resolved through mathematical formalism alone, instead requiring deliberation about the politics of development as well as the context of deployment. Drawing from a new sociotechnical lexicon, we redefine vagueness in terms of distinct design challenges at key stages in AI system development. The resulting framework of Hard Choices in Artificial Intelligence (HCAI) empowers developers by 1) identifying points of overlap between design decisions and major sociotechnical challenges; 2) motivating the creation of stakeholder feedback channels so that safety issues can be exhaustively addressed. As such, HCAI contributes to a timely debate about the status of AI development in democratic societies, arguing that deliberation should be the goal of AI Safety, not just the procedure by which it is ensured.
Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate co... more Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate complex dynamical systems with nonlinear dynamics for the purpose of analysis, design and control. Motivated by inference of spatio-temporal dynamics in breast cancer research, we propose a method to efficiently solve an identification problem for a specific class of discrete-time LTV systems, in which the states are fully observed and there is no access to system inputs. In addition, it is assumed that we do not know on which states the inputs act, which can change between time steps, and that the total number of inputs is sparse over all states and over time. The problem is formulated as a compressive sensing problem, which incorporates the effect of measurement noise and which has a solution with a partially sparse support. We derive sufficient conditions for the unique recovery of the system model and input values, which lead to practical conditions on the number of experiments and rank conditions on system outputs. Synthetic experiments analyze the method's sensitivity to noise for randomly generated models.
Real-Time Distribution Grid State Estimation with Limited Sensors and Load Forecasting
High penetration levels of distributed generation (DG) and electric vehicles (EVs) diversify powe... more High penetration levels of distributed generation (DG) and electric vehicles (EVs) diversify power flow and bring uncertainty to distribution networks, making planning and control more involved for distribution system operators (DSOs). The increased risk of constraint violation triggers the need to augment forecasts with real- time state estimation. This is economically and technically challenging since it requires investing in a large number of sensors and these have to communicate with often older and slower supervisory control and data acquisition (SCADA) systems. We address distribution grid state estimation via combining only a limited set of sensors with load forecast information. It revisits open problems in a recent paper that proposes a Bayesian estimation scheme. We derive the estimator for balanced power networks via rigorous modeling. An off-line analysis of load aggregation, forecast accuracy and number of sensors provides concrete engineering trade-offs to determine the optimal number of sensors for a desired accuracy. This estimation procedure can be used in real time as an observer for control problems or off-line for planning purposes to asses the effect of DG or EVs on specific network components.
Implementing state estimation in low and medium voltage power distribution is still challenging g... more Implementing state estimation in low and medium voltage power distribution is still challenging given the scale of many networks and the reliance of traditional methods on a large number of measurements. This paper proposes a method to improve voltage predictions in real-time by leveraging a limited set of real-time measurements. The method relies on Bayesian estimation formulated as a linear least squares estimation problem, which resembles the classical weighted least-squares (WLS) approach for scenarios where full network observability is not available. We build on recently developed linear approximations for unbalanced three-phase power flow to construct voltage predictions as a linear mapping of load predictions constructed with Gaussian processes. The estimation step to update the voltage forecasts in real-time is a linear computation allowing fast high-resolution state estimate updates. The uncertainty in forecasts can be determined a priori and smoothed a posteriori, making the method useful for both planning, operation and posthoc analysis. The method outperforms conventional WLS and is applied to different test feeders and validated on a real test feeder with the utility Alliander in The Netherlands.
This paper provides an empirical and conceptual account on seeing machine learning models as part... more This paper provides an empirical and conceptual account on seeing machine learning models as part of a sociotechnical system to identify relevant vulnerabilities emerging in the context of use. As ML is increasingly adopted in socially sensitive and safety-critical domains, many ML applications end up not delivering on their promises, and contributing to new forms of algorithmic harm. There is still a lack of empirical insights as well as conceptual tools and frameworks to properly understand and design for the impact of ML models in their sociotechnical context. In this paper, we follow a design science research approach to work towards such insights and tools. We center our study in the financial industry, where we first empirically map recently emerging MLOps practices to govern ML applications, and corroborate our insights with recent literature. We then perform an integrative literature research to identify a long list of vulnerabilities that emerge in the sociotechnical context of ML applications, and we theorize these along eight dimensions. We then perform semi-structured interviews in two real-world use cases and across a broad set of relevant actors and organizations, to validate the conceptual dimensions and identify challenges to address sociotechnical vulnerabilities in the design and governance of ML-based systems. The paper proposes a set of guidelines to proactively and integrally address both the dimensions of sociotechnical vulnerability, as well as the challenges identified in the empirical use case research, in the organization of MLOps practices.
Frontiers in artificial intelligence and applications, Jun 22, 2023
Values, such as freedom and safety, are the core motivations that guide us humans. A prerequisite... more Values, such as freedom and safety, are the core motivations that guide us humans. A prerequisite for creating value-aligned multiagent systems that involve humans and artificial agents is value inference, the process of identifying values and reasoning about human value preferences. We introduce a framework that connects the value inference steps, and motivate why a hybrid intelligence approach is instrumental for its success. We also highlight the multidisciplinary research challenges that hybrid value inference entails.
My supervisor Alessandro Abate has introduced me to the exciting and challenging field of Systems... more My supervisor Alessandro Abate has introduced me to the exciting and challenging field of Systems Biology, and has provided me with the unconditional freedom, to pursue my own ideas in the project while learning extensively from his valuable experiences. He has been steadily available to give me guidance and inspiration, from the moment we met in Berkeley in 2008, till this day. It has been a great pleasure to work with him. Bart De Schutter, my professor at DCSC, has inspired me to aim for the highest during his lectures, and has helped me to pursue my ambitions by providing the contacts and the support, which were necessary to enable this experience. Claire Tomlin, my professor at EECS, has inspired me tremendously, and has offered a great deal of effective feedback to set the right directions in my research. I would like to thank Peter Wieringa and Paul van den Hof for their support during the early stage of my Master. Jeff Axelrod, at the Stanford School of Medicine, has been a great collaborator, with his critical eye and his many comprehensible introductions in the biological domain. Stephane Vincent has been of great help by providing the biological knowledge and hypotheses, and by interpreting my work from a biological perspective. Alberto Silletti has contributed to this project in first person by providing his computer science expertise, and his useful feedback on how to structure data. Lots of my growth as a MSc student can be attributed to the interaction with fellow students and researchers. I would like to thank the Control Boys at Delft, i.e. Lodewijk Alblas, Robert Hillen, Vincent Gusdorf, and Riny Vermue, for our many fruitful discussions, and cheerful moments at 3mE. Moreover, I send a word of gratitude to the people of the Hybrid Systems Lab in Berkeley for their hospitality, in particular to Insoon Yang, Young-Hwan Chang, and Pranav Shah, and most memorable to Soulaiman Itani, who was steadily available to give me advise, and help me out where needed. Neal Master has contributed significantly to the project through his extracurricular devotion, and abundant curiosity. Indirectly, I have benefited a lot from some exceptional people at Berkeley. I want to thank Adam Arkin, Alexandre Bayen, and Murat Arcak for allowing me to follow their lectures. I am grateful to my aunt Gloria and my uncle Fred, for making me feel at home in California. A number of people in and around the International House has made my Berkeley experience unforgettable. Thanks to all of you, and specifically to my friends Brit Moller, Amaury de Closset, and Pablo Calabuig.
As AI systems are integrated into high stakes social domains, researchers now examine how to desi... more As AI systems are integrated into high stakes social domains, researchers now examine how to design and operate them in a safe and ethical manner. However, the criteria for identifying and diagnosing safety risks in complex social contexts remain unclear and contested. In this paper, we examine the vagueness in debates about the safety and ethical behavior of AI systems. We show how this vagueness cannot be resolved through mathematical formalism alone, instead requiring deliberation about the politics of development as well as the context of deployment. Drawing from a new sociotechnical lexicon, we redefine vagueness in terms of distinct design challenges at key stages in AI system development. The resulting framework of Hard Choices in Artificial Intelligence (HCAI) empowers developers by 1) identifying points of overlap between design decisions and major sociotechnical challenges; 2) motivating the creation of stakeholder feedback channels so that safety issues can be exhaustively addressed. As such, HCAI contributes to a timely debate about the status of AI development in democratic societies, arguing that deliberation should be the goal of AI Safety, not just the procedure by which it is ensured.
Toward Distributed Energy Services: Decentralizing Optimal Power Flow With Machine Learning
IEEE Transactions on Smart Grid, Mar 1, 2020
The implementation of optimal power flow (OPF) methods to perform voltage and power flow regulati... more The implementation of optimal power flow (OPF) methods to perform voltage and power flow regulation in electric networks is generally believed to require extensive communication. We consider distribution systems with multiple controllable Distributed Energy Resources (DERs) and present a data-driven approach to learn control policies for each DER to reconstruct and mimic the solution to a centralized OPF problem from solely locally available information. Collectively, all local controllers closely match the centralized OPF solution, providing near-optimal performance and satisfaction of system constraints. A rate distortion framework enables the analysis of how well the resulting fully decentralized control policies are able to reconstruct the OPF solution. The methodology provides a natural extension to decide what nodes a DER should communicate with to improve the reconstruction of its individual policy. The method is applied on both single- and three-phase test feeder networks using data from real loads and distributed generators, focusing on DERs that do not exhibit intertemporal dependencies. It provides a framework for Distribution System Operators to efficiently plan and operate the contributions of DERs to achieve Distributed Energy Services in distribution networks.
IEEE/ACM Transactions on Computational Biology and Bioinformatics, Jul 1, 2016
With the growth of high-throughput proteomic data, in particular time series gene expression data... more With the growth of high-throughput proteomic data, in particular time series gene expression data from various perturbations, a general question that has arisen is how to organize inherently heterogenous data into meaningful structures. Since biological systems such as breast cancer tumors respond differently to various treatments, little is known about exactly how these gene regulatory networks (GRNs) operate under different stimuli. Challenges due to the lack of knowledge not only occur in modeling the dynamics of a GRN but also cause bias or uncertainties in identifying parameters or inferring the GRN structure. This paper describes a new algorithm which enables us to estimate bias error due to the effect of perturbations and correctly identify the common graph structure among biased inferred graph structures. To do this, we retrieve common dynamics of the GRN subject to various perturbations. We refer to the task as "repairing" inspired by "image repairing" in computer vision. The method can automatically correctly repair the common graph structure across perturbed GRNs, even without precise information about the effect of the perturbations. We evaluate the method on synthetic data sets and demonstrate an application to the DREAM data sets and discuss its implications to experiment design.
Real-time data-driven optimization and control problems over networks, such as in traffic or ener... more Real-time data-driven optimization and control problems over networks, such as in traffic or energy systems, may require sensitive information of participating agents to calculate solutions and decision variables. Adversaries with access to coordination signals may potentially decode information on individual agents and put privacy at risk. We use the Inexact Alternating Minimization Algorithm to instantiate local differential privacy for distributed optimization, addressing situations in which individual agents need to protect their individual data, in the form of optimization parameters, from all other agents and any central authority. This mechanism allows agents to customize their own privacy level based on local needs and parameter sensitivities. The resulting algorithm works across a large family of convex distributed optimization problems. We implement the method on a distributed optimal power flow problem that aims to prevent overload on critical branches in a radial network.
IEEE/ACM Transactions on Computational Biology and Bioinformatics, Nov 1, 2012
Epithelia are sheets of connected cells that are essential across the animal kingdom. Experimenta... more Epithelia are sheets of connected cells that are essential across the animal kingdom. Experimental observations suggest that the dynamical behavior of many single-layered epithelial tissues has strong analogies with that of specific mechanical systems, namely large networks consisting of point masses connected through spring-damper elements and undergoing the influence of active and dissipating forces. Based on this analogy, this work develops a modeling framework to enable the study of the mechanical properties and of the dynamic behavior of large epithelial cellular networks. The model is built first by creating a network topology that is extracted from the actual cellular geometry as obtained from experiments, then by associating a mechanical structure and dynamics to the network via spring-damper elements. This scalable approach enables running simulations of large network dynamics: the derived modeling framework in particular is predisposed to be tailored to study general dynamics (for example, morphogenesis) of various classes of single-layered epithelial cellular networks. In this contribution we test the model on a case study of the dorsal epithelium of the Drosophila melanogaster embryo during early dorsal closure (and, less conspicuously, germband retraction).
A growing body of experimental evidence indicates a strong link between intratumoral heterogeneit... more A growing body of experimental evidence indicates a strong link between intratumoral heterogeneity and therapeutic resistance in cancer. In particular, tumor cells may survive therapy by switching their phenotypic identities to more resistant, drug-tolerant states. Computational models of phenotypic plasticity in response to cytotoxic therapy are needed: (1) to strengthen understanding of the interplay between phenotypic heterogeneity and therapeutic resistance, and (2) to identify potential strategies in silico that weaken resistance prior to in vitro testing. This work presents a linear time-invariant model of phenotypic state dynamics to deduce subpopulation-level behavior likely to affect temporal phenotypic composition and thus drug resistance. The model was identified under different therapeutic conditions with authentic biological data from a breast cancer cell line. Subsequent analysis suggested drug-induced effects on phenotypic state switching that could not be deduced directly from empirical observations. A bootstrap algorithm was implemented to identify statistically significant results: reduction in cell division under each therapeutic condition versus control. Further, Monte Carlo simulation was used to evaluate quality of model fit for twoway switching and net switching on synthetically generated data to determine the limitations of the latter assumption for subsequent modeling. Most importantly, the simple model structure initiated a control-theoretic approach for identifying promising combination treatments in silico to guide future laboratory testing.
Learning cooperative policies for multi-agent systems is often challenged by partial observabilit... more Learning cooperative policies for multi-agent systems is often challenged by partial observability and a lack of coordination. In some settings, the structure of a problem allows a distributed solution with limited communication. Here, we consider a scenario where no communication is available, and instead we learn local policies for all agents that collectively mimic the solution to a centralized multi-agent static optimization problem. Our main contribution is an information theoretic framework based on rate distortion theory which facilitates analysis of how well the resulting fully decentralized policies are able to reconstruct the optimal solution. Moreover, this framework provides a natural extension that addresses which nodes an agent should communicate with to improve the performance of its individual policy.
As AI systems become prevalent in high stakes domains such as surveillance and healthcare, resear... more As AI systems become prevalent in high stakes domains such as surveillance and healthcare, researchers now examine how to design and implement them in a safe manner. However, the potential harms caused by systems to stakeholders in complex social contexts and how to address these remains unclear. In this paper, we explain the inherent normative uncertainty in debates about the safety of AI systems. We then address this as a problem of vagueness by examining its place in the design, training, and deployment stages of AI system development. We adopt Ruth Chang's theory of intuitive comparability to illustrate the dilemmas that manifest at each stage. We then discuss how stakeholders can navigate these dilemmas by incorporating distinct forms of dissent into the development pipeline, drawing on Elizabeth Anderson's work on the epistemic powers of democratic institutions. We outline a framework of sociotechnical commitments to formal, substantive and discursive challenges that address normative uncertainty across stakeholders, and propose the cultivation of related virtues by those responsible for development.
This paper reframes algorithmic systems as intimately connected to and part of social and ecologi... more This paper reframes algorithmic systems as intimately connected to and part of social and ecological systems, and proposes a first-ofits-kind methodology for environmental justice-oriented algorithmic audits. How do we consider environmental and climate justice dimensions of the way algorithmic systems are designed, developed, and deployed? These impacts are inherently emergent and can only be understood and addressed at the level of relations between an algorithmic system and the social (including institutional) and ecological components of the broader ecosystem it operates in. As a result, we claim that in absence of an integral ontology for algorithmic systems, we cannot do justice to the emergent nature of broader environmental impacts of algorithmic systems and their underlying computational infrastructure. Furthermore, an integral lens provides many lessons from the history of environmental justice that are of relevance in current day struggles for algorithmic justice. We propose to define algorithmic systems as ontologically indistinct from Social-Ecological-Technological Systems (SETS), framing emergent implications as couplings between social, ecological, and technical components of the broader fabric in which algorithms are integrated and operate. We draw upon prior work on SETS analysis as well as emerging themes in the literature and practices of Environmental Justice (EJ) to conceptualize and assess algorithmic impact. We then offer three policy recommendations to help establish a SETS-based EJ approach to algorithmic audits: (1) broaden the inputs and open-up the outputs of an audit, (2) enable meaningful access to redress, and (3) guarantee a place-based and relational approach to the process of evaluating impact. We operationalize these as a qualitative framework of questions for a spectrum of stakeholders. Doing so, this article aims to inspire stronger and more frequent interactions across policymakers, researchers, practitioners, civil society, and grassroots communities.
Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate co... more Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate complex dynamical systems with nonlinear dynamics for the purpose of analysis, design and control. Motivated by inference of spatio-temporal dynamics in breast cancer research, we propose a method to efficiently solve an identification problem for a specific class of discrete-time LTV systems, in which the states are fully observed and there is no access to system inputs. In addition, it is assumed that we do not know on which states the inputs act, which can change between time steps, and that the total number of inputs is sparse over all states and over time. The problem is formulated as a compressive sensing problem, which incorporates the effect of measurement noise and which has a solution with a partially sparse support. We derive sufficient conditions for the unique recovery of the system model and input values, which lead to practical conditions on the number of experiments and rank conditions on system outputs. Synthetic experiments analyze the method's sensitivity to noise for randomly generated models.
As AI systems are integrated into high stakes social domains, researchers now examine how to desi... more As AI systems are integrated into high stakes social domains, researchers now examine how to design and operate them in a safe and ethical manner. However, the criteria for identifying and diagnosing safety risks in complex social contexts remain unclear and contested. In this paper, we examine the vagueness in debates about the safety and ethical behavior of AI systems. We show how this vagueness cannot be resolved through mathematical formalism alone, instead requiring deliberation about the politics of development as well as the context of deployment. Drawing from a new sociotechnical lexicon, we redefine vagueness in terms of distinct design challenges at key stages in AI system development. The resulting framework of Hard Choices in Artificial Intelligence (HCAI) empowers developers by 1) identifying points of overlap between design decisions and major sociotechnical challenges; 2) motivating the creation of stakeholder feedback channels so that safety issues can be exhaustively addressed. As such, HCAI contributes to a timely debate about the status of AI development in democratic societies, arguing that deliberation should be the goal of AI Safety, not just the procedure by which it is ensured.
Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate co... more Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate complex dynamical systems with nonlinear dynamics for the purpose of analysis, design and control. Motivated by inference of spatio-temporal dynamics in breast cancer research, we propose a method to efficiently solve an identification problem for a specific class of discrete-time LTV systems, in which the states are fully observed and there is no access to system inputs. In addition, it is assumed that we do not know on which states the inputs act, which can change between time steps, and that the total number of inputs is sparse over all states and over time. The problem is formulated as a compressive sensing problem, which incorporates the effect of measurement noise and which has a solution with a partially sparse support. We derive sufficient conditions for the unique recovery of the system model and input values, which lead to practical conditions on the number of experiments and rank conditions on system outputs. Synthetic experiments analyze the method's sensitivity to noise for randomly generated models.
Real-Time Distribution Grid State Estimation with Limited Sensors and Load Forecasting
High penetration levels of distributed generation (DG) and electric vehicles (EVs) diversify powe... more High penetration levels of distributed generation (DG) and electric vehicles (EVs) diversify power flow and bring uncertainty to distribution networks, making planning and control more involved for distribution system operators (DSOs). The increased risk of constraint violation triggers the need to augment forecasts with real- time state estimation. This is economically and technically challenging since it requires investing in a large number of sensors and these have to communicate with often older and slower supervisory control and data acquisition (SCADA) systems. We address distribution grid state estimation via combining only a limited set of sensors with load forecast information. It revisits open problems in a recent paper that proposes a Bayesian estimation scheme. We derive the estimator for balanced power networks via rigorous modeling. An off-line analysis of load aggregation, forecast accuracy and number of sensors provides concrete engineering trade-offs to determine the optimal number of sensors for a desired accuracy. This estimation procedure can be used in real time as an observer for control problems or off-line for planning purposes to asses the effect of DG or EVs on specific network components.
Implementing state estimation in low and medium voltage power distribution is still challenging g... more Implementing state estimation in low and medium voltage power distribution is still challenging given the scale of many networks and the reliance of traditional methods on a large number of measurements. This paper proposes a method to improve voltage predictions in real-time by leveraging a limited set of real-time measurements. The method relies on Bayesian estimation formulated as a linear least squares estimation problem, which resembles the classical weighted least-squares (WLS) approach for scenarios where full network observability is not available. We build on recently developed linear approximations for unbalanced three-phase power flow to construct voltage predictions as a linear mapping of load predictions constructed with Gaussian processes. The estimation step to update the voltage forecasts in real-time is a linear computation allowing fast high-resolution state estimate updates. The uncertainty in forecasts can be determined a priori and smoothed a posteriori, making the method useful for both planning, operation and posthoc analysis. The method outperforms conventional WLS and is applied to different test feeders and validated on a real test feeder with the utility Alliander in The Netherlands.
This paper provides an empirical and conceptual account on seeing machine learning models as part... more This paper provides an empirical and conceptual account on seeing machine learning models as part of a sociotechnical system to identify relevant vulnerabilities emerging in the context of use. As ML is increasingly adopted in socially sensitive and safety-critical domains, many ML applications end up not delivering on their promises, and contributing to new forms of algorithmic harm. There is still a lack of empirical insights as well as conceptual tools and frameworks to properly understand and design for the impact of ML models in their sociotechnical context. In this paper, we follow a design science research approach to work towards such insights and tools. We center our study in the financial industry, where we first empirically map recently emerging MLOps practices to govern ML applications, and corroborate our insights with recent literature. We then perform an integrative literature research to identify a long list of vulnerabilities that emerge in the sociotechnical context of ML applications, and we theorize these along eight dimensions. We then perform semi-structured interviews in two real-world use cases and across a broad set of relevant actors and organizations, to validate the conceptual dimensions and identify challenges to address sociotechnical vulnerabilities in the design and governance of ML-based systems. The paper proposes a set of guidelines to proactively and integrally address both the dimensions of sociotechnical vulnerability, as well as the challenges identified in the empirical use case research, in the organization of MLOps practices.
Frontiers in artificial intelligence and applications, Jun 22, 2023
Values, such as freedom and safety, are the core motivations that guide us humans. A prerequisite... more Values, such as freedom and safety, are the core motivations that guide us humans. A prerequisite for creating value-aligned multiagent systems that involve humans and artificial agents is value inference, the process of identifying values and reasoning about human value preferences. We introduce a framework that connects the value inference steps, and motivate why a hybrid intelligence approach is instrumental for its success. We also highlight the multidisciplinary research challenges that hybrid value inference entails.
My supervisor Alessandro Abate has introduced me to the exciting and challenging field of Systems... more My supervisor Alessandro Abate has introduced me to the exciting and challenging field of Systems Biology, and has provided me with the unconditional freedom, to pursue my own ideas in the project while learning extensively from his valuable experiences. He has been steadily available to give me guidance and inspiration, from the moment we met in Berkeley in 2008, till this day. It has been a great pleasure to work with him. Bart De Schutter, my professor at DCSC, has inspired me to aim for the highest during his lectures, and has helped me to pursue my ambitions by providing the contacts and the support, which were necessary to enable this experience. Claire Tomlin, my professor at EECS, has inspired me tremendously, and has offered a great deal of effective feedback to set the right directions in my research. I would like to thank Peter Wieringa and Paul van den Hof for their support during the early stage of my Master. Jeff Axelrod, at the Stanford School of Medicine, has been a great collaborator, with his critical eye and his many comprehensible introductions in the biological domain. Stephane Vincent has been of great help by providing the biological knowledge and hypotheses, and by interpreting my work from a biological perspective. Alberto Silletti has contributed to this project in first person by providing his computer science expertise, and his useful feedback on how to structure data. Lots of my growth as a MSc student can be attributed to the interaction with fellow students and researchers. I would like to thank the Control Boys at Delft, i.e. Lodewijk Alblas, Robert Hillen, Vincent Gusdorf, and Riny Vermue, for our many fruitful discussions, and cheerful moments at 3mE. Moreover, I send a word of gratitude to the people of the Hybrid Systems Lab in Berkeley for their hospitality, in particular to Insoon Yang, Young-Hwan Chang, and Pranav Shah, and most memorable to Soulaiman Itani, who was steadily available to give me advise, and help me out where needed. Neal Master has contributed significantly to the project through his extracurricular devotion, and abundant curiosity. Indirectly, I have benefited a lot from some exceptional people at Berkeley. I want to thank Adam Arkin, Alexandre Bayen, and Murat Arcak for allowing me to follow their lectures. I am grateful to my aunt Gloria and my uncle Fred, for making me feel at home in California. A number of people in and around the International House has made my Berkeley experience unforgettable. Thanks to all of you, and specifically to my friends Brit Moller, Amaury de Closset, and Pablo Calabuig.
As AI systems are integrated into high stakes social domains, researchers now examine how to desi... more As AI systems are integrated into high stakes social domains, researchers now examine how to design and operate them in a safe and ethical manner. However, the criteria for identifying and diagnosing safety risks in complex social contexts remain unclear and contested. In this paper, we examine the vagueness in debates about the safety and ethical behavior of AI systems. We show how this vagueness cannot be resolved through mathematical formalism alone, instead requiring deliberation about the politics of development as well as the context of deployment. Drawing from a new sociotechnical lexicon, we redefine vagueness in terms of distinct design challenges at key stages in AI system development. The resulting framework of Hard Choices in Artificial Intelligence (HCAI) empowers developers by 1) identifying points of overlap between design decisions and major sociotechnical challenges; 2) motivating the creation of stakeholder feedback channels so that safety issues can be exhaustively addressed. As such, HCAI contributes to a timely debate about the status of AI development in democratic societies, arguing that deliberation should be the goal of AI Safety, not just the procedure by which it is ensured.
Toward Distributed Energy Services: Decentralizing Optimal Power Flow With Machine Learning
IEEE Transactions on Smart Grid, Mar 1, 2020
The implementation of optimal power flow (OPF) methods to perform voltage and power flow regulati... more The implementation of optimal power flow (OPF) methods to perform voltage and power flow regulation in electric networks is generally believed to require extensive communication. We consider distribution systems with multiple controllable Distributed Energy Resources (DERs) and present a data-driven approach to learn control policies for each DER to reconstruct and mimic the solution to a centralized OPF problem from solely locally available information. Collectively, all local controllers closely match the centralized OPF solution, providing near-optimal performance and satisfaction of system constraints. A rate distortion framework enables the analysis of how well the resulting fully decentralized control policies are able to reconstruct the OPF solution. The methodology provides a natural extension to decide what nodes a DER should communicate with to improve the reconstruction of its individual policy. The method is applied on both single- and three-phase test feeder networks using data from real loads and distributed generators, focusing on DERs that do not exhibit intertemporal dependencies. It provides a framework for Distribution System Operators to efficiently plan and operate the contributions of DERs to achieve Distributed Energy Services in distribution networks.
IEEE/ACM Transactions on Computational Biology and Bioinformatics, Jul 1, 2016
With the growth of high-throughput proteomic data, in particular time series gene expression data... more With the growth of high-throughput proteomic data, in particular time series gene expression data from various perturbations, a general question that has arisen is how to organize inherently heterogenous data into meaningful structures. Since biological systems such as breast cancer tumors respond differently to various treatments, little is known about exactly how these gene regulatory networks (GRNs) operate under different stimuli. Challenges due to the lack of knowledge not only occur in modeling the dynamics of a GRN but also cause bias or uncertainties in identifying parameters or inferring the GRN structure. This paper describes a new algorithm which enables us to estimate bias error due to the effect of perturbations and correctly identify the common graph structure among biased inferred graph structures. To do this, we retrieve common dynamics of the GRN subject to various perturbations. We refer to the task as "repairing" inspired by "image repairing" in computer vision. The method can automatically correctly repair the common graph structure across perturbed GRNs, even without precise information about the effect of the perturbations. We evaluate the method on synthetic data sets and demonstrate an application to the DREAM data sets and discuss its implications to experiment design.
Real-time data-driven optimization and control problems over networks, such as in traffic or ener... more Real-time data-driven optimization and control problems over networks, such as in traffic or energy systems, may require sensitive information of participating agents to calculate solutions and decision variables. Adversaries with access to coordination signals may potentially decode information on individual agents and put privacy at risk. We use the Inexact Alternating Minimization Algorithm to instantiate local differential privacy for distributed optimization, addressing situations in which individual agents need to protect their individual data, in the form of optimization parameters, from all other agents and any central authority. This mechanism allows agents to customize their own privacy level based on local needs and parameter sensitivities. The resulting algorithm works across a large family of convex distributed optimization problems. We implement the method on a distributed optimal power flow problem that aims to prevent overload on critical branches in a radial network.
IEEE/ACM Transactions on Computational Biology and Bioinformatics, Nov 1, 2012
Epithelia are sheets of connected cells that are essential across the animal kingdom. Experimenta... more Epithelia are sheets of connected cells that are essential across the animal kingdom. Experimental observations suggest that the dynamical behavior of many single-layered epithelial tissues has strong analogies with that of specific mechanical systems, namely large networks consisting of point masses connected through spring-damper elements and undergoing the influence of active and dissipating forces. Based on this analogy, this work develops a modeling framework to enable the study of the mechanical properties and of the dynamic behavior of large epithelial cellular networks. The model is built first by creating a network topology that is extracted from the actual cellular geometry as obtained from experiments, then by associating a mechanical structure and dynamics to the network via spring-damper elements. This scalable approach enables running simulations of large network dynamics: the derived modeling framework in particular is predisposed to be tailored to study general dynamics (for example, morphogenesis) of various classes of single-layered epithelial cellular networks. In this contribution we test the model on a case study of the dorsal epithelium of the Drosophila melanogaster embryo during early dorsal closure (and, less conspicuously, germband retraction).
A growing body of experimental evidence indicates a strong link between intratumoral heterogeneit... more A growing body of experimental evidence indicates a strong link between intratumoral heterogeneity and therapeutic resistance in cancer. In particular, tumor cells may survive therapy by switching their phenotypic identities to more resistant, drug-tolerant states. Computational models of phenotypic plasticity in response to cytotoxic therapy are needed: (1) to strengthen understanding of the interplay between phenotypic heterogeneity and therapeutic resistance, and (2) to identify potential strategies in silico that weaken resistance prior to in vitro testing. This work presents a linear time-invariant model of phenotypic state dynamics to deduce subpopulation-level behavior likely to affect temporal phenotypic composition and thus drug resistance. The model was identified under different therapeutic conditions with authentic biological data from a breast cancer cell line. Subsequent analysis suggested drug-induced effects on phenotypic state switching that could not be deduced directly from empirical observations. A bootstrap algorithm was implemented to identify statistically significant results: reduction in cell division under each therapeutic condition versus control. Further, Monte Carlo simulation was used to evaluate quality of model fit for twoway switching and net switching on synthetically generated data to determine the limitations of the latter assumption for subsequent modeling. Most importantly, the simple model structure initiated a control-theoretic approach for identifying promising combination treatments in silico to guide future laboratory testing.
Learning cooperative policies for multi-agent systems is often challenged by partial observabilit... more Learning cooperative policies for multi-agent systems is often challenged by partial observability and a lack of coordination. In some settings, the structure of a problem allows a distributed solution with limited communication. Here, we consider a scenario where no communication is available, and instead we learn local policies for all agents that collectively mimic the solution to a centralized multi-agent static optimization problem. Our main contribution is an information theoretic framework based on rate distortion theory which facilitates analysis of how well the resulting fully decentralized policies are able to reconstruct the optimal solution. Moreover, this framework provides a natural extension that addresses which nodes an agent should communicate with to improve the performance of its individual policy.
As AI systems become prevalent in high stakes domains such as surveillance and healthcare, resear... more As AI systems become prevalent in high stakes domains such as surveillance and healthcare, researchers now examine how to design and implement them in a safe manner. However, the potential harms caused by systems to stakeholders in complex social contexts and how to address these remains unclear. In this paper, we explain the inherent normative uncertainty in debates about the safety of AI systems. We then address this as a problem of vagueness by examining its place in the design, training, and deployment stages of AI system development. We adopt Ruth Chang's theory of intuitive comparability to illustrate the dilemmas that manifest at each stage. We then discuss how stakeholders can navigate these dilemmas by incorporating distinct forms of dissent into the development pipeline, drawing on Elizabeth Anderson's work on the epistemic powers of democratic institutions. We outline a framework of sociotechnical commitments to formal, substantive and discursive challenges that address normative uncertainty across stakeholders, and propose the cultivation of related virtues by those responsible for development.
This paper reframes algorithmic systems as intimately connected to and part of social and ecologi... more This paper reframes algorithmic systems as intimately connected to and part of social and ecological systems, and proposes a first-ofits-kind methodology for environmental justice-oriented algorithmic audits. How do we consider environmental and climate justice dimensions of the way algorithmic systems are designed, developed, and deployed? These impacts are inherently emergent and can only be understood and addressed at the level of relations between an algorithmic system and the social (including institutional) and ecological components of the broader ecosystem it operates in. As a result, we claim that in absence of an integral ontology for algorithmic systems, we cannot do justice to the emergent nature of broader environmental impacts of algorithmic systems and their underlying computational infrastructure. Furthermore, an integral lens provides many lessons from the history of environmental justice that are of relevance in current day struggles for algorithmic justice. We propose to define algorithmic systems as ontologically indistinct from Social-Ecological-Technological Systems (SETS), framing emergent implications as couplings between social, ecological, and technical components of the broader fabric in which algorithms are integrated and operate. We draw upon prior work on SETS analysis as well as emerging themes in the literature and practices of Environmental Justice (EJ) to conceptualize and assess algorithmic impact. We then offer three policy recommendations to help establish a SETS-based EJ approach to algorithmic audits: (1) broaden the inputs and open-up the outputs of an audit, (2) enable meaningful access to redress, and (3) guarantee a place-based and relational approach to the process of evaluating impact. We operationalize these as a qualitative framework of questions for a spectrum of stakeholders. Doing so, this article aims to inspire stronger and more frequent interactions across policymakers, researchers, practitioners, civil society, and grassroots communities.
Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate co... more Discrete-time linear time-varying (LTV) systems form a powerful class of models to approximate complex dynamical systems with nonlinear dynamics for the purpose of analysis, design and control. Motivated by inference of spatio-temporal dynamics in breast cancer research, we propose a method to efficiently solve an identification problem for a specific class of discrete-time LTV systems, in which the states are fully observed and there is no access to system inputs. In addition, it is assumed that we do not know on which states the inputs act, which can change between time steps, and that the total number of inputs is sparse over all states and over time. The problem is formulated as a compressive sensing problem, which incorporates the effect of measurement noise and which has a solution with a partially sparse support. We derive sufficient conditions for the unique recovery of the system model and input values, which lead to practical conditions on the number of experiments and rank conditions on system outputs. Synthetic experiments analyze the method's sensitivity to noise for randomly generated models.
Uploads
Papers by Roel Dobbe