Christina Fragouli - US grants

This project aims to advance our fundamental understanding of secrecy over arbitrary wireless networks. Over the last decade we have significantly deepened our fundamental understanding on how to send information over wireless networks, while our understanding on how to securely send this information has not reached the same depth as yet. This project aims to develop a unifying theory that enhances wireless secrecy by exploiting wireless properties such as: the existence of feedback (today part of all wireless standards); the possibility of selecting and using multiple network communication paths; the smart use of wireless jamming and the wireless channel variability and unpredictability. We enable this by "building on erasures:" through appropriate coding and smart wireless jamming, we convert (Gaussian) wireless networks to erasure networks. We then develop protocols that use interaction and feedback to enable provable security against active and passive adversaries, even if they are computationally unbounded.

The project also promotes the training of research engineers: we will integrate the research into the curriculum via the creation of novel coursework combining the underlying concepts in wireless communication, network coding/protocols, and information theory. The research in this project, if successful, will contribute to the fundamental sciences of information theoretic network security and secure network coding. Given our increasing dependence on wireless devices as a portal for socio-economic activities, wireless security will have broad implications in mobile commerce.

Wireless access is fast becoming the primary portal to the Internet, causing an exponentially rising demand for wireless data. This has pushed current wireless systems to their limits, despite significant investment in infrastructure to meet the ever-growing demands. Physical layer cooperation can enable near-optimal usage of the available wireless bandwidth.

This proposal fundamentally rethinks physical-layer cooperation by introducing an on-demand approach to wireless cooperation. It can be argued that the success of the (wired) Internet was made possible by its on-demand operation, using adaptation, local knowledge, feedback and a best-effort service model. Notably, the success of many peer-to-peer network protocols is closely tied to on-demand information propagation and adaptation of the network. The broad goal of this proposal is to bring this philosophy to wireless networks, enabling near-optimal usage of the wireless network bandwidth within the complexity constraints of implementable systems. This project aims to complement the theoretical work with proof-of-concept deployments on software radio testbeds, and also engage industry partners to impact next-generation wireless network designs. The project also promotes training of research engineers, through a plan to establish a unique inter-university education and research program, which will include joint and collaborative student advising and curricular development.

The underlying assumption of most network information theory works is that one can build architectures which tightly coordinate the estimation and sharing of information about the wireless channels, the user requirements and the network topology, at a very fast time-scale, without impacting the performance (rates, error). That is, the cost of learning the very dynamic network state is not accounted for. Another implicit assumption is complete network usage, that all available relays in a wireless network are used, with no adaptation to user demand. This can be very wasteful in many situations, and again, the cost of unnecessarily using relays is not accounted for. The many breakthrough ideas based on these assumptions have advanced a collective understanding, but bringing physical layer cooperation techniques closer to practical networks requires additional steps to move beyond these assumptions. This project puts together a program that develops the theoretical foundations and practice of an on-demand network operation, that dispenses with these assumptions. This entails operating specific subsets of the network relays (sub-networks) that fulfill target rates, as opposed to using all network relays to achieve the best possible performance. This project constructs a theoretical understanding of how to select, adapt and operate these sub-networks on demand, by using accountable partial network knowledge and using feedback mechanisms to enhance signal adaptation to the unknown. The theoretical formulations are tightly coupled to implementable protocols that will be validated in test beds.

Coding has traditionally aimed to securely and efficiently convey specific information messages, for example speech in a phone call, to one or more receivers. Today?s communication networks, however, increasingly deliver content rather than a specific message. Popular networks that serve content-type traffic include advertising networks, news aggregators, and social media. This proposal formulates a novel theoretical framework aiming to determine the fundamental performance limits and design principles for content-type networks. This new research direction promotes the progress of science and has the potential to transform the way content-type traffic is encoded and transmitted in networks. As a result, this research is expected to benefit society at large by laying the foundation for a more efficient design of network services, and to be of immediate and far-reaching use for both private and public sectors. This proposal integrates discovery research and education: it develops a rich learning experience for students through an integrated inter-university program and curricular development. The investigators will keep serving as a role model for minority students, especially young women, by active engagement at their institutions and within their peer societies.

The research work in this proposal is fundamentally motivated by current trends in the Internet traffic: today society increasingly relies on content-type services for educational, professional and social purposes. This proposal explores coding specifically targeted to content-type communication; it can therefore form a solid theoretical foundation for the design and optimization of future network services. This proposal formulates new research directions not explored before by addressing the following main technical objectives: (1) Develop an information theoretical framework by proposing a novel model for classical message-specific index coding and extending it to content-type pliable index coding, deriving inner and outer bounds, and efficient numerical algorithms for their evaluation. (2) Derive fundamental bounds and practical coding strategies for content-type coding, where clients are interested in multiple, as opposed to a single content, identify order approximations and study extensions to noisy network versions of the problem. (3) Develop oblivious strategies for servers with limited knowledge about the clients' side-information (for instance, the server knows the number of downloads of the clients but not necessarily which ones); quantify the amount and quality of the side-information necessary for optimal performance. (4) Investigate the case where all content is not "equal" (for instance, because of personalized recommendations based on the client's past behavior); connect with recommendation systems, and understand how preferences affect the content-type algorithms and bounds.

For each technical objective, the investigators will develop tools from information theory, network coding, and algorithms to address the novel questions posed in this research. The new framework is expected to significantly advance the state-of-the-art and fundamental understanding of content-type coding from both a theoretical (information theory) and a more practical (code and algorithm design) perspective. The results of this research will be timely presented at major national and international professional venues in the information theory, network coding, and networking communities. The proposal has also a strong educational and outreach component. In order to reach the general public the investigators will leverage the departmental media tools to demonstrate the developed technology and its practical impact. Students will receive a solid foundation in communications, coding and networking thus acquiring the fundamental skills to be successful in the competitive, diverse, and global workforce market. It is the investigators? goal to serve as a role model for minority students, especially women.

Consistent with the National Science Foundation's goal "to initiate and support ... programs to strengthen scientific and engineering research potential [and] science and engineering education programs at all levels", this project develops materials that will support education and broad awareness of the importance of key engineering advances . In particular, it will support creation of educational films and corresponding lesson materials for K-12 mathematics and science teachers that will allow students to be exposed not just to the advances in information theory, but also to how an ordinary person played a pivotal role in fostering them. By adapting material previously restricted to graduate-level courses and technical conferences to a larger audience, a broader dissemination of information theory should profitably inform teachers and researchers in other fields, thereby fueling the nation's STEM workforce and improving commercial technology.

This project focuses on the efforts and advances of Claude Shannon, who established the field of information theory by stating some of its most fundamental problems and solving them. The technologies that his work made possible form a major driving force of our economy. His information theory concepts provide the foundation for nearly every aspect of modern information technology and have been applied to many fields, including communication, language, genetics, computing, cryptography, psychology, perception, memory, artificial intelligence, quantum physics, and others.

This project develops a novel Cyber Physical System (CPS) centric approach to privacy and security for wireless networked CPS systems, by reconciling the low-delay and low-jitter requirements of CPS applications with the requirements imposed by security and privacy. Our starting observation is that, in CPS, an adversary's primary goal is not to learn all the raw data, but instead core attributes, such as the state or control actions that are derived from data. Building on this observation, we propose to use a distortion measure for security that maximizes the difference between the eavesdropper's estimate and the true value of the function computing the attributes of interest, reducing the adversary's ability to disrupt normal operation of CPS. We posit that we can protect these core attributes with fewer resources than needed to protect all the raw data. Ensuring secure and private information exchange over networked CPS systems is essential to building a thriving ecosystem of applications that range from autonomous cars and drones, to the Internet-of-Things (IoT), to immersive environments such as augmented reality for health, education, and collaboration. Our educational plan engages not only graduate students and postdocs but also high school and undergraduate students. It also reaches out to engineers and the lay public, by providing open source implementations of our algorithms making them available both to industry and hobbyists.

The project considers both passive and active attacks. We will quantify novel privacy and security measures for CPS systems that are based on distortion measurements in a metric space; we will develop fundamental bounds as well as low complexity and low overhead coding schemes; we will quantify the disruptive power of active adversaries and design pro-active and retro-active defense mechanisms; and we will illustrate our approach over a flagship application, drone localization. Our approach will offer an alternative to wireless network encryption methods, by designing for low-delay, low-jitter requirements of CPS.

Wireless communications and networks have experienced exponential growth in data rates and traffic over the past decade, driven by the ever-increasing density of mobile devices, multimedia services and data requirements. The resulting electromagnetic spectrum below 60-GHz has become extremely overcrowded, even with advanced spectrum-efficient modulation formats and spatially diverse multiple-input multiple-output (MIMO) techniques. At present, the sub-millimeter-wave (sub-mm-Wave) electromagnetic spectrum between 300 GHz and 850 GHz is largely unassigned and provides a unique opportunity for more efficient utilization. This will avoid further crowding the currently heavily used spectrum and significantly enhance data rates to tens of Gb/s. This project seeks to demonstrate such a fundamentally new platform towards spectral-efficient and energy-efficient wireless communications with embedded security. Due to the inherent atmospheric attenuation, the sub-mm-Wave communication distance has been limited to within 50 m. Thus, this project proposes network configurations of sub-mm-Wave point-to-point links to enable secured spatial coverage over longer distances and larger areas. There are two distinct differences of the sub-mm-Wave links compared to traditional wireless networks: the directivity of the sub-mm-Wave links and the possibility for a transmitter to connect to multiple receivers through adaptive electronic beam-steering and beam-forming. The beam-forming with narrow beam-width removes broadcasting and avoids interference, enabling much simpler network operation to approach the theoretical upper limits of network information capacity. The project seeks to demonstrate the modular sub-mm-Wave link hardware to achieve the above goal. The proposed research will be complemented with an integrated education and outreach program. This includes diversity recruitment, mentoring and retention, hands-on curriculum development, minority high-school and undergraduate training, and public outreach. The cross-layer scientific and education provides a new platform at the interface of hardware, software, and networks in next-generation wireless communication networks.

This project will develop a spectrally dense, high-data-rate, 650-GHz photonic wireless communications platform in a diamond mesh network, while explicitly addressing network security and energy efficiency in the architecture. The collaborative research spans across the physical layer, the network layer, and the software layer, addressing cross-layer issues in the fundamental architecture. The proposed research consists of three thrust areas. In Thrust I, the project will examine a modular photonic sub-mm-Wave link, based on a chip-scale photomixer driven by an optical frequency comb recently developed by the team. This enables high-power spectrally dense, 80 Gb/s sub-mm-Wave transmission. In Thrust II, the project will examine a photonic sub-mm-Wave 80 Gb/s testbed, implemented with an adaptive smart antenna array. Beam-steering and beam-forming will enable simultaneously a directional line-of-sight (LoS) link and a non-line-of-sight (NLoS) link, with the former establishing a spectrally efficient channel with less inter-symbol and inter-channel interference. The latter mitigates medium non-idealities such as interference, shadowing, and multi-path effects. In Thrust III, the project will study the capacity of sub-mm-Wave communication networks and explore the design of near-optimal efficient and secure algorithms. Enabled by the intrinsic directivity and beam-forming capabilities of our sub-mm-Wave link, the project will advance the possibility of unconditional security in the wireless backhaul network through physical layer security algorithms.

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.

Data compression is a core component of all communication protocols, as it can translate to bandwidth savings, energy efficiency and low delay operations. In the traditional setup, an information source compresses its messages so that they can be communicated efficiently with the goal of ensuring accurate reconstruction at the destination. This project seeks to design compression schemes that are specifically tailored to Machine Learning applications: If the transmitted messages support a given learning task (e.g., classification or learning), the desired compression schemes should provide better support for the learning task instead of focusing on reconstruction accuracy. This approach to compression could potentially yield significant benefits in terms of communication efficiency, while simultaneously promoting the successful implementation of Machine Learning algorithms. By improving communication efficiency, such schemes are expected to contribute to the successful implementation of distributed machine learning algorithms over networks.

Traditionally, compression schemes are evaluated using rate-distortion trade-offs; this project is interested in rate-accuracy trade-offs, where accuracy captures the effect that quantization may have on a specific machine learning task. There is particular interest in information-theoretic lower bounds and trade-offs, and in explicit compression for the following two questions: (1) How to compress for model training, when we need to use distributed communication constrained nodes to learn a model, fast and efficiently; and (2) How to compress for communication during inference. The project will derive bounds and algorithms for distributed compression of features coming from composite distributions that will be used for a machine learning task, such as classification. This work will advance the state of the art, and build new connections between the areas of data compression and distributed machine learning.

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.

Active learning and online learning are machine-learning paradigms in which computers learn to make complex decisions while receiving feedback from an environment. For instance, a drone may learn to fly by itself, or a car may learn to drive by trial and error. Recently, these learning paradigms have been widely applied and have achieved phenomenal successes with human-level performance in tasks like gameplay or robot control. As computing devices become smaller and less power-consuming, new distributed learning frameworks start to emerge. These frameworks contain low-capability learning agents (such as cell phones, unmanned vehicles, or drones) that are far apart but perform learning collectively by communicating with each other through (wireless) networks. However, existing communication approaches would become bottlenecks for learning since they were designed for high-power computers and consume too much power and network bandwidth. This project aims to address this issue by providing novel techniques that efficiently compress data to be communicated while preserving the learning ability. The techniques developed in this project will advance the state-of-the-art in distributed online/active learning by improving communication efficiencies. <br/><br/>The overarching goal of this project is to establish efficient compression schemes that support effective active/online learning, such as bandit and reinforcement learning over communication-constrained networks. In these learning environments, a learner aims to make a good decision for the next steps based on experience; this project will explore fundamental bounds and efficient algorithms that support this goal while minimizing the number of bits communicated - by compressing in a way that only retains the necessary information for decision making. In other words, this project aims to explore the fundamental trade-off between compression and learnability in active/online environments. Building on promising preliminary work, the investigators will study problems ranging from the most basic multi-arm bandit setting to more complex reinforcement learning settings and consider both centralized and decentralized network topologies. More specifically, the investigators propose compression schemes and fundamental theoretical bounds for (1) rewards in multi-armed bandit problems, (2) context vectors for contextual bandit problems, and (3) state-action features and models for Markov decision problems.<br/><br/>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.