Alejandro Ribeiro - US grants

The objective of this research is to develop the scientific foundation for the quantitative analysis and design of control networks. Control networks are wireless substrates for industrial automation control, such as the WirelessHART and Honeywell's OneWireless, and have fundamental differences over their sensor network counterparts as they also include actuation and the physical dynamics. The approach of the project focuses on understanding cross-cutting interfaces between computing systems, control systems, sensor networks, and wireless communications using time-triggered architectures.

The intellectual merit of this research is based on using time-triggered communication and computation as a unifying abstraction for understanding control networks. Time-triggered architectures enable the natural integration of communication, computation, and physical aspects of control networks as switched control systems. The time-triggered abstraction will serve for addressing the following interrelated themes: Optimal Schedules via Quantitative Automata, Quantitative Analysis and Design of Control Networks: Wireless Protocols for Optimal Control: Quantitative Trust Management for Control Networks.

Various components of this research will be integrated into the PIs' RAVEN control network which is compatible with both WirelessHART and OneWireless specifications. This provides a direct path for this proposal to have immediate industrial impact. In order to increase the broader impact of this project, this project will launch the creation of a Masters' program in Embedded Systems, one of the first in the nation. The principle that guides the curriculum development of this novel program is a unified systems view of computing, communication, and control systems.

The last decades of the 20th century witnessed the emergence of networks that have become ever more pervasive and important. At the start of the 21st century developing a scientific understanding of networks remains an unresolved intellectual endeavor. Part of this effort is the understanding of wireless communication networks on a fundamental level pursued in this project. The research formulates optimization problems that model wireless networks in a variety of settings and translates properties of the former into characteristics of the latter. For doing so, the investigators rely on the fact that randomness, in the form of fading, yields seemingly more complex problems that nonetheless have a more tractable structure. This richer structure is studied to: (i) Determine architectural properties of wireless networks. (ii) Design algorithms to find optimal operating points for different types of physical layers. (iii) Develop strategies to learn fading distributions. (iv) Consider tradeoffs associated with acquisition of channel state information.

The education agenda revolves around the excitement, challenge and discipline gaps. The excitement gap is about the excitement people feels about science and technology versus the lack of interest to pursue careers in science and technology. The challenge gap refers to the ongoing trend to reduce the complexity of material covered in courses. The discipline gap alludes to the compartmental experience offered to students and the reality of an increasingly hazy separation between disciplines. The education plan contributes to the closing of these gaps through the development of an undergraduate level course on stochastic processes and a graduate level course on optimal design of wireless networks. Both of these courses are designed to be challenging, exciting and multidisciplinary.

Abstract

Sensor networks are interactive collections of distributed devices that interface the virtual and physical worlds. Among the tasks needed to implement these interfaces is the inference of properties of the physical world using observations collected by the network. Inference tasks cover a vast range, particular examples being estimation of rainfall in an orchard, or tracking the surface salinity of the ocean.
This project develops an integrated framework for distributed statistical inference of dynamic processes using sensor networks. The ultimate goal is to impact on the numerous activities that stand to benefit from the development of sensor networks, including economic sectors like manufacture, agriculture, and environmental management.
Further impact is expected from the incorporation of research results in undergraduate classes.

This project advances the use of prices to mediate the incorporation of global knowledge into local estimates. Many problems in dynamic statistical inference require solution of optimization problems, prompting formulations whereby estimates are viewed as economic outputs to be maximized. The global optimization that would result from the action of a centralized agent is regarded as a social resource optimization problem. Local estimates computed by individual sensors are the result of selfish actions of market agents. Prices are introduced to align social and agents? interest. While these ideas have been successfully pursued in static environments, this project pursues them in dynamic settings. The research cuts horizontally across different dynamic statistical inference problems and is vertically organized into:
(i) Determination of convergence properties of price mediation algorithms. (ii) Resolution of memory growth problems through manipulation of price structures. (iii) Practical considerations including robustness, convergence speed, and communication effects (iv) Integration with learning algorithms for problems with incomplete model information.

This project develops an axiomatic theory of hierarchical clustering for asymmetric networks as is typical of trust propagation. Say, for example, that Miranda trusts Billy who trusts Ariel who trusts Miranda, but there has not been enough interactions in the opposite direction to establish trust. When these three people meet, shall they trust each other? The answer to this question is equivalent to the determination of whether Miranda, Billy, and Ariel are part of the same cluster: their circle of trust. The axioms of value, influence, and transformation are postulated. The Axiom of Value says that in a two-node network the nodes cluster at resolution equal to the maximum dissimilarity between them. The Axiom of Influence says that no clusters are formed at resolutions that do not allow bidirectional paths to be formed. The Axiom of Transformation states that if we consider a network and reduce all pairwise dissimilarities, the level at which two nodes become part of the same cluster is not larger than the level at which they were clustered together in the original network. Generic properties of any method that satisfies these axioms are explored and specific methods that abide by these axioms are derived. To enrich the axiomatic exploration of asymmetric clustering an application thrust and three theory thrusts are pursued. The application thrust explores the formation of circles of trust in social networks and the design of protocols to establish trust in technological networks. The theory thrusts will study alternative axiomatic formulations, stability of asymmetric hierarchical clustering algorithms, and the determination of algorithms to compute hierarchical clusters.

The educational agenda is integrated into the Market and Social Systems Engineering (MKSE) program at the University of Pennsylvania. The MKSE program is an undergraduate course of study that fully integrates the disciplines needed to design and analyze the complex networks that are reshaping our society. Given the importance of trust in these networks the research undertaken in the context of this project is incorporated into classes in the MKSE program. The excitement an idea like the formalization of trust propagation generates is further exploited to draw attention to the MKSE and Systems Engineering programs from the wider academic community. These ideas are part of a long term effort on the part of the PI to contribute to the closing of the excitement, challenge, and discipline gaps.

The goal of this project is to develop a new framework where teams of mobile robots self-engineer the structure of their communication network in order to improve on the performance of distributed algorithms used for team coordination. The key idea that enables this work is the representation of the network structure in terms of metrics that depend on the full eigenvalue spectrum of the network's adjacency and Laplacian matrices, but can be approximated in a very efficient and decentralized way. These metrics can then be related to the performance of popular, distributed, coordination algorithms, such as consensus, gossiping, and viral information dissemination. The intellectual merit of this research lies in the development of a hybrid network of mobile robots that is controlled jointly in the space of network configurations and robot positions. The study of the integrated system requires the synthesis of new theoretical results drawing from control theory, spectral graph theory, wireless networking, and optimization. This hybrid network combines the following interrelated objectives: Spectral analysis and distributed control of robot networks; Integrated network and mobility control; Richer models of the communication space; Platform deployment and validation.

Successful completion of this research will provide these necessary components in facilitating the design of mobile autonomous systems and fostering their adoption. Wide availability of such systems can have a significant societal impact on, e.g., search, rescue and recovery operations, environmental monitoring for homeland security, or surveillance and reconnaissance missions. The broader impact of this project lies on disseminating the research output in the industry and academia.

Network data, defined as one that encodes relationships between elements, is a pervasive component of modern data analytics. The purpose of this project is to advance our capacity to process and understand network data. The contention is that the central difficulty in analyzing large-scale complex networks comes from lack of structure. This loose nature contrasts with the rigidity of a closely related construction: the metric space. If understanding networks is challenging but understanding metric spaces is not, a route to network analysis is to project networks into metric spaces. This motivates the technical goal of designing methods and algorithms to implement these projections.

The project builds on preliminary results combining a projection axiom (networks that are already metric remain unchanged after projection) and a dissimilarity-reducing axiom (smaller networks have smaller projections) to establish existence and uniqueness results. Further explorations are pursued in four thrusts: (1) Study of projection methods in symmetric networks. (2) Incorporation of asymmetric networks and asymmetric quasimetric spaces. (3) Metric representations derived from triangle inequalities written in dioid algebras; which albeit metric in an abstract sense are very different from regular metrics. (4) Generalizations to high order networks in which dissimilarities are defined for tuples other than binary. Applications to search, network comparison, and diffusion processes, complement the theoretical research. Broader impacts come through industrial partnerships and an aggressive educational agenda that leverages the University of Pennsylvania's institutional commitment to play a leading role in the education of engineers that are to exploit the opportunities afforded by the ever increasing access to data and the ever broadening scope that networks play in our society.

The internet-of-things (IoT) revolution is bringing millions of physical devices online (e.g. cars, UAVs, homes, medical devices), enabling them to connect to each other in real-time, as well as to cloud services. Wireless communication will be critical in providing IoT connectivity. There are three main strategic directions that are envisioned in the future wireless networking. First, enhanced mobile broadband will result in even larger data rates for future 3D applications such as virtual reality. Second, Smart City/Community applications require large number of sensors that communicate sporadically over large urban or rural areas in a scalable, asynchronous, and energy efficient manner. The previous two directions, while important, are not the focus of our proposal. Instead, our proposal focuses on low-latency and ultra-reliable communications and networking that is critical for latency-sensitive, closed-loop control applications, like vehicle to vehicle communications, collaborative swam planning, and industrial control. In such latency sensitive applications, we need to rethink the networking stack, coding, networking architecture, and control design to enable communications and networking that can provide ultra-low latency (99.999%). This is far beyond what is currently possible. But even more importantly, we do not know what is possible and what are the fundamental limits for control system design over low-latency, high-reliability communications.

In this proposal, we will be rethinking the scientific foundations for ultra-reliable, low-latency wireless communications for latency sensitive control applications. We propose to achieve our scientific agenda by addressing three intellectual challenges: 1) Low-latency channel coding, where the goal is to focus on short packet codes for control loops 2) Control over low latency-aware communication channels, where the goal is to understand the what is the optimal tradeoff of latency to reliability for control loops and 3) Learning for Large Scale Wireless Control Networks, where machine learning will perform resource allocation for large numbers of control loops with competing latency/reliability requirements We intend to evaluate the proposed research agenda by leveraging our existing Intel Science and Technology Center (ISTC) on Wireless Autonomous Systems and demonstrate our ideas in future wireless protocols (IEEE 802.11ax) and experimentally demonstrate it in high-speed V2V and fast formation control with aerial swarms. On the educational front, the University of Pennsylvania is planning to offer a Micro-Masters program in Internet of Things (IoT) on the edX MOOC platform. Longer term, our goal is to create a new community of researchers that focus on control over low-latency wireless networks for IoT devices. Towards this goal, we plan on leveraging departmental efforts to increase and diversify the PhD students working on this project.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Recent advances in artificial intelligence have led to significant progress in our ability to extract information from images and time sequences. Maintaining this rate of progress hinges upon attaining equally significant results in the processing of more complex signals such as those that are acquired by autonomous systems and networks of connected devices, or those that arise in the study of complex biological and social systems. This award establishes FINPenn, the Center for the Foundations of Information Processing at the University of Pennsylvania. The focus of the center is to establish fundamental theory to enable the study of data beyond time and images. The center's premise is that humans' rich intuitive understanding of space and time may not necessarily be applicable to the processing of complex signals. Therefore, matching the success in time and space necessitates the discovery and development of foundational principles to guide the design of generic artificial intelligence algorithms. FINPenn will support a class of scholar trainees along with a class of visiting postdocs and students to advance this agenda. The center will engage the community through the organization of workshops and lectures and will disseminate knowledge with onsite and online educational activities at the undergraduate and graduate level.

FINPenn builds on two observations: (i) To understand the foundations of data science it is necessary to succeed beyond Euclidean signals in time and space. This is true even to understand the foundations for Euclidean signal processing. (ii) Humans live in Euclidean time and space. To succeed in information processing beyond signals with Euclidean structure, operation from foundational principles is necessary because human intuition is of limited help. For instance, convolutional neural networks have found success in the processing of images and signals in time but they rely heavily on spatial and temporal intuition. To generalize their success to unconventional signal domains it is necessary to postulate fundamental principles and generalize from those principles. If the generalizations are successful they not only illuminate the new application domains but they also help establish the validity of the postulated principles for Euclidean spaces in the tradition of predictive science. The proposers further contend that the foundational principles of data sciences are to be found in the exploitation of structure and the associated invariances and symmetries that structure generates. The initial focus of the center is in advancing the theory of information processing in signals whose structure is defined by a group, a graph, or a topology. These three types of signals generate three foundational research directions which build on the particular strengths of the University of Pennsylvania on network sciences, robotics, and autonomous systems which are areas in which these types of signals appear often.

This project is part of the National Science Foundation's Harnessing the Data Revolution (HDR) Big Idea activity.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Recent advances in deep learning have led to many disruptive technologies: from automatic speech recognition systems, to automated supermarkets, to self-driving cars. However, the complex and large-scale nature of deep networks makes them hard to analyze and, therefore, they are mostly used as black-boxes without formal guarantees on their performance. For example, deep networks provide a self-reported confidence score, but they are frequently inaccurate and uncalibrated, or likely to make large mistakes on rare cases. Moreover, the design of deep networks remains an art and is largely driven by empirical performance on a dataset. As deep learning systems are increasingly employed in our daily lives, it becomes critical to understand if their predictions satisfy certain desired properties. The goal of this NSF-Simons Research Collaboration on the Mathematical and Scientific Foundations of Deep Learning is to develop a mathematical, statistical and computational framework that helps explain the success of current network architectures, understand its pitfalls, and guide the design of novel architectures with guaranteed confidence, robustness, interpretability, optimality, and transferability. This project will train a diverse STEM workforce with data science skills that are essential for the global competitiveness of the US economy by creating new undergraduate and graduate programs in the foundations of data science and organizing a series of collaborative research events, including semester research programs and summer schools on the foundations of deep learning. This project will also impact women and underrepresented minorities by involving undergraduates in the foundations of data science.

Deep networks have led to dramatic improvements in the performance of pattern recognition systems. However, the mathematical reasons for this success remain elusive. For instance, it is not clear why deep networks generalize or transfer to new tasks, or why simple optimization strategies can reach a local or global minimum of the associated non-convex optimization problem. Moreover, there is no principled way of designing the architecture of the network so that it satisfies certain desired properties, such as expressivity, transferability, optimality and robustness. This project brings together a multidisciplinary team of mathematicians, statisticians, theoretical computer scientists, and electrical engineers to develop the mathematical and scientific foundations of deep learning. The project is divided in four main thrusts. The analysis thrust will use principles from approximation theory, information theory, statistical inference, and robust control to analyze properties of deep networks such as expressivity, interpretability, confidence, fairness and robustness. The learning thrust will use principles from dynamical systems, non-convex and stochastic optimization, statistical learning theory, adaptive control, and high-dimensional statistics to design and analyze learning algorithms with guaranteed convergence, optimality and generalization properties. The design thrust will use principles from algebra, geometry, topology, graph theory and optimization to design and learn network architectures that capture algebraic, geometric and graph structures in both the data and the task. The transferability thrust will use principles from multiscale analysis and modeling, reinforcement learning, and Markov decision processes to design and study data representations that are suitable for learning from and transferring to multiple tasks.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.