Federated Learning (FL) is a machine learning method for training with private data locally stored in distributed machines without gathering them into one place for central learning. Despite its promises, FL is prone to critical security risks. First, because FL depends on a central server to aggregate local training models, this is a single point of failure. The server might function maliciously. Second, due to its distributed nature, FL might encounter backdoor attacks by participating clients. They can poison the local model before submitting to the server. Either type of attack, on the server or the client side, would severely degrade learning accuracy. We propose FedBlock, a novel blockchain-based FL framework that addresses both of these security risks. FedBlock is uniquely desirable in that it involves only smart contract programming, thus deployable atop any blockchain network. Our framework is substantiated with a comprehensive evaluation study using real-world datasets. Its robustness against backdoor attacks is competitive with the literature of FL backdoor defense. The latter, however, does not address the server risk as we do.

Compared to typical crypto-exchange AMMs, fixed-rate AMMs are much more difficult to design due to the inter-dependency of interest rate, loan price, and, especially, time factor. The literature lacks scientific developments. Existing efforts are either too simplistic, focusing more on automation and less on economic effect, or too engineering-driven, using ad hoc formulations without economic justifications. The fact is that today’s fixed-rate DeFi is struggling to attract any side of participation, whether it is the borrowers, the lenders, and the liquidity providers (LPs). We propose BondMM, a new fixed-rate AMM algorithm as an effort to advance the state of the art. It has the mechanics of a constant-function AMM with path-independence property, which is desirable for avoiding money-pumping attacks. It is constructed, whose correctness provable, using elegant closed-form mathematics, hence efficient for blockchain implementation. The properties and performance of BondMM are validated with our theoretical analysis and simulation-based evaluation. Specifically, we show that today’s benchmark fixed-rate AMMs are either too financially-risky (Yield AMM) or too trader-expensive (Notional AMM) to sustain practically for the long term. BondMM is superior to both regarding properties that define the financial strength of a lending institution: 1) better price impact to attract borrowers and lenders, 2) better capital efficiency to attract LPs, and 3) better financial stability for both participants and LPs.

As mobile devices have become ubiquitous, traffic at the edge of the network is growing faster than ever. To improve user experience, commodity servers are deployed in the edge to form a distributed network of mini datacenters. A consequential task is to partition the user cells into groups, each to be served by an edge server, to maximize the offloading to the edge. Most often this task can be cast as a graph partitioning problem, where a graph is used to represent the workload and the goal of the partition is to satisfy a given objective, for example, minimum cut. This approach is effective for a fixed workload. In contrast, we address the case where the workload changes over the time. We propose a probabilistic partitioning technique that is more robust to workload dynamics. The novelty of this technique is in the representation of the time-varying workload as a random variable of a probability distribution. We evaluate our theory with an evaluation using real-world datasets.

The geometric embedding approach to network dissimilarity estimation assumes a linear relationship between the dissimilarity and the intrinsic metric. In this paper, we conjecture that this relationship should better be formulated by a power law. We propose a geometric framework based on this conjecture and show that by choosing the right power coefficient one can achieve better estimation accuracy. This is substantiated in an evaluation study using both synthetic and real-world datasets.

Network latency estimation is a long-studied problem. A well-accepted approach is to construct a network coordinate system whose distance is used as an approximation for the unknown latency. As the accuracy depends on the geometry underlying this system, it is important to find the best fitting geometric space. We propose a geometric framework that enables investigation of any kind of such a space. We also provide efficient algorithms for Euclidean and Riemannian embeddings which represent the most widely used geometric spaces. The proposed theory is substantiated in an evaluation using real-world datasets.

Data partitioning is a crucial component of any distributed storage system that wants to scale. For retrieval efficiency, data frequently requested together in the same query should be placed on the same server as much as possible. Although intuitive, this is not easy to be implemented if constrained by load balancing; computationally, it is an NP hard problem. Existing research has offered approximate solutions optimized for a given workload of queries, in which the order as to when each query is received is not considered. This paper initiates a new study on online partitioning algorithms that are sequentially optimized for a query sequence. In the new problem, the queries arrive in a stream manner, unknown, and given the option to revise the partition after each query, the objective is to minimize the total query processing cost and data migration cost. We formulate this problem formally, investigate several online heuristics, and evaluate them using simulation.

Location fingerprinting is a range-free approach to GPS-free localization. Conventionally, the fingerprint space is defined as a feature vector space where a fingerprint is a vector of location-sensitive measurements associated with a location. However, in practice, it is hard to find a quality feature space that is robust to device heterogeneity and environment and infrastructure dynamics. This paper advocates a fundamentally different model where a fingerprint is defined as a dissimilarity measurement associated with a pair of locations and proposes a localization approach based on geometric embedding.

This paper initiates a new study on online partitioning algorithms that are sequentially optimized for a query sequence. As queries arrive one at a time, given the option to reconfigure the partition after each query so that it can best serve the next query, the objective is to minimize the query read cost and data migration cost. This is an online problem without an optimal solution; online heuristics are the only resort. We investigate this problem by formulating it as a multi-objective optimization and proposing an Evolutionary Algorithms (EA) framework incorporating several online heuristics to explore Pareto-optimal solutions. This study is driven by our conjecture that if an online heuristic helps EA converge faster to better partitioning solutions then practically this heuristic should be preferred for adapting the partition to the query sequence.

Location fingerprinting is a common approach to indoor localization. For good accuracy, the training set of sample fingerprints should be sufficiently large to be well-representative of the environment in terms of both spatial coverage and temporal coverage. As such, the computation required during the positioning phase can be expensive because we have to evaluate each new fingerprint against the training data repeatedly over time. It is desirable, therefore, to optimize computational efficiency, not just localization accuracy. Existing techniques are far from this goal due to their polarization toward one criterion but not both. We propose a substantially better technique based on the novel approach of modeling indoor localization as a classification learning problem where classes form a spatial hierarchy. Its performance is substantiated in our evaluation study.

We address the problem of fingerprint-based tracking of a moving device where training fingerprints are not available all at once but arrive sporadically in a stream manner. As such the only information that may be used to locate the device is its current fingerprint and the set of fingerprints obtained earlier impromptu. The challenge is to devise a localization algorithm that can scale with the fingerprint stream as this stream can grow limitlessly with time. We propose an online algorithm framework that requires storage of and computation based on only a constant-size sliding buffer of fingerprints, not all the fingerprints in the past. Our algorithm, therefore, is fast and efficient. In terms of localization error, it offers a good approximation when compared to the expensive batch algorithm that assumes access to the entire history of fingerprints.

Much research has been done in the last several years on social network dynamics as online social networks have become ubiquitous. While networks like Facebook and LinkedIn have given the research world access to data on large-scale, general human social interaction in an online environment, there are a growing number of smaller-scale social networks aimed toward certain niche user groups which can provide a richer context for the study of social network dynamics. One such network is Currensee, an emerging social network for foreign currency exchange (Forex) traders. Since all of the trades represented on the Currensee network are the actual trades being made by users, it can serve as a good model for the study of financial activity by individuals in the context of a social network. We present in this paper an analysis of this network in terms of its connectivity, social activity, and trading activity.

Human dynamics in real-world social networks has been a long lasting topic of research. Recently, the rapid growing of online social networks with their data made accessible has given researchers a door of opportunity to study large-scale social networks. Facebook has been the largest, most influential, and fastest growing social network on the Internet to date. As Facebook's userbase is extremely diverse, spreading across different countries, races, ages, professions, and interests, to name a few, this network is a good model to study our society. In this paper, we present an analysis of a regional Facebook dataset, with attention given to the correlation among user connectivity, activity, and similarity.

In data-centric sensor networks, sensor data is not necessarily forwarded to a central sink for storage; instead, the nodes themselves serve as a distributed in-network storage, collectively storing sensor data and waiting to answer user queries. A key problem in designing such a network is how to map data and queries to their corresponding rendezvous nodes so that a query can find its matching data quickly and efficiently. Existing techniques are mostly aimed to address a certain type of queries. The capability to support queries of any type is desirable, yet remains a challenge. We propose a search- efficient architecture for data-centric sensor networks, that can simultaneously address range queries and top-k, the two most popular query types. Importantly, the proposed technique does not require location information, hence applicable to a wide range of sensor networks. Our theoretical findings are complemented by a simulation-based evaluation.

Maintenance is an important issue in distributed storage systems. In many cases, such as in archival storage where data access is infrequent, the workload due to maintaining a system is dominant and much heavier than that to answering retrieval queries. Erasure coding, widely used in many distributed storage systems for its high reliability, is efficient for maintenance but not so efficient for access. In this paper, we investigate an extension of this method with the purpose of improving maintenance efficiency and provide an analysis on the tradeoff of this extension.

We propose CRES as a substrate structure to enable content-based routing in large-scale sensor networks without location information. With CRES, traffic is routed based on its content rather than the destination's physical address. CRES is therefore applicable to data-centric sensor networks where sensor data and user queries need to be mapped to their corresponding nodes in a distributed manner such that a query can find its matching data efficiently.

We propose a new mechanism for generating networks with a wide variety of degree distributions. The idea is a modification of the well-studied preferential attachment scheme in which the degree of each node is used to determine its evolving connectivity. Modifications to this base protocol to include features other than connectivity have been considered in building the network. However, schemes based on preferential attachment in any form require substantial information on the entire network. We propose instead a protocol based only on a single statistical feature which results from the reasonable assumption that the effect of various attributes, which determine the ability of each node to attract other nodes, is multiplicative. This composite attribute or fitness is lognormally distributed and is used in forming the complex network. We show that, by varying the parameters of the lognormal distribution, we can recover both exponential and power-law degree distributions. The exponents for the power-law case are in the correct range seen in real- world networks such as the World Wide Web and the Internet. Further, as power-law networks with exponents in the same range are a crucial ingredient of efficient search algorithms in peer-to- peer networks, we believe our network construct may serve as a basis for new protocols that will enable peer-to-peer networks to efficiently establish a topology conducive to optimized search procedures.

Target tracking techniques usually assume that the target is within the detection range of one or more sensor nodes at known locations, so that the target's location can be determined based on these locations. Instead, we are interested in networks where most sensor nodes do not know their location; hence, it is most likely that the target is many hops away from the closest location-known sensors. For such networks, we propose a target localization system that can estimate the target location based on hop-count information only. We evaluate the performance of this system via simulation.

CAN is a well-known DHT technique for content- based P2P networks, where each node is assigned a zone in a virtual coordinate space to store the index of the data hashed into this zone. The dimension of this space is usually lower than the data dimension, thus we have the problem of dimension mismatch. This problem is widely addressed in the context of data retrieval that follows the traditional request/response model. However, little has been done for the publish/subscribe model, which is the focus of our paper. We show that dimension mismatch in CAN-based publish/subscribe applications poses new challenges. We furthermore investigate how a random projection approach can help reduce the negative effects of dimension mismatch. Our theoretical findings are complemented by a simulation-based evaluation.

This paper summarizes a key finding in our NSF project CNS-0615055/0753066. The project is aimed at development and implementation of a complete architectural design for information retrieval in decentralized networks. In this paper, we particularly focus on two main components: the communication architecture and the indexing architecture. The communication component serves as the foundation for the nodes to maintain connectivity in the system, while the indexing component prepares the information in a way that enables fast and efficient future search operations. A challenge is to overcome the curse of dimensionality. From both network and database perspectives, decentralized networks are complex; not only we have to accommodate the rapid growth of the network size, but also the multi-dimensionality of the data space. We describe our solution in the context of P2P networks, but it is also applicable to any Internet-based decentralized networks.

Publish/subscribe services are natural applications of sensor networks as sensors are designed mainly to detect and notify upon events of interest. To enable these services, the brokerage mechanism that routes a given event to the subscribing nodes must be confined to the networkpsilas resource constraints. GHT has been used as a cost-effective way to disseminate subscriptions and events to the sensor network. It is, however, efficient only for exact-match subscriptions. We propose an efficient GHT-like technique for both exact-match and range subscriptions where the data space can be of any dimension.

In recent years, the number of Peer-to-Peer (P2P) applications has increased significantly. One important problem in many P2P applications is how to efficiently disseminate data from a single source to multiple receivers on the Internet. A successful model used for analyzing this problem is a graph consisting of nodes and edges, with a capacity assigned to each edge. In some situations however, it is inconvenient to use this model. To that end, we propose to study the problem of efficient data dissemination in asource constraintnetwork. A source constraint network is modeled as a graph in which, the capacity is associated with a node, rather than an edge. The contributions of this paper include (a) a quantitative data dissemination in any source constraint network, (b) a set of topologies suitable for data dissemination in P2P networks, and (c) an architecture and implementation of a P2P system based on the proposed optimal topologies. We will present the experimental results of our P2P system deployed on PlanetLab nodes demonstrating that our approach achieves near optimal throughput while providing scalability, low delay and bandwidth fairness among peers.

We consider the problem of estimating the geographic locations of nodes in a wireless sensor network where most sensors are without an effective self-positioning functionality. We propose LSVM-a novel solution with the following merits. First, LSVM localizes the network based on mere connectivity information (that is, hop counts only) and therefore is simple and does not require specialized ranging hardware or assisting mobile devices as in most existing techniques. Second, LSVM is based on Support Vector Machine (SVM) learning. Although SVM is a classification method, we show its applicability to the localization problem and prove that the localization error can be upper bounded by any small threshold given an appropriate training data size. Third, LSVM addresses the border and coverage-hole problems effectively. Last but not least, LSVM offers fast localization in a distributed manner with efficient use of processing and communication resources. We also propose a modified version of mass-spring optimization to further improve the location estimation in LSVM. The promising performance of LSVM is exhibited by our simulation study.

Subscription covering detection is useful to improving the performance of any publish/subscribe system. However, an exact solution to querying coverings among a large set of subscriptions in high dimension is computationally too expensive to be practicable. Therefore, we are interested in an approximate approach. We focus on spherical subscriptions and propose a solution based on random projections. Our complexities are substantially better than that of the exact approach. The proposed solution can potentially find exact coverings with a success probability 100% asymptotically approachable.

Mobility, channel error, and congestion are the main causes for packet loss in mobile ad hoc networks. Reducing packet loss typically involves congestion control operating on top of a mobility and failure adaptive routing protocol at the network layer. In the current designs, routing is not congestion-adaptive. Routing may let a congestion happen which is detected by congestion control, but dealing with congestion in this reactive manner results in longer delay and unnecessary packet loss and requires significant overhead if a new route is needed. This problem becomes more visible especially in large-scale transmission of heavy traffic such as multimedia data, where congestion is more probable and the negative impact of packet loss on the service quality is of more significance. We argue that routing should not only be aware of, but also be adaptive to, network congestion. Hence, we propose a routing protocol (CRP) with such properties. Our ns-2 simulation results confirm that CRP improves the packet loss rate and end-to-end delay while enjoying significantly smaller protocol overhead and higher energy efficiency as compared to AODV and DSR

Mobility, channel error, and congestion are the main causes for packet loss in mobile ad hoc networks. Reducing packet loss typically involves congestion control operating on top of a mobility and failure adaptive routing protocol at the network layer. In current designs, routing is not congestion-adaptive. Routing may let congestion happen, which is detected by congestion control, but dealing with congestion in this reactive manner results in longer delay and unnecessary packet loss and requires significant overhead if a new route is needed. This problem becomes especially more visible in large-scale transmission of heavy traffic such as multimedia data, where congestion is more probable and the negative impact of packet loss on the service quality is more significant. We argue that routing should not only be aware of, but also be adaptive to, network congestion. We propose a routing protocol with such properties.

We present a design for a system that provides video-on-demand (VOD) services to mobile ad hoc clients. Such a system allows the clients to access video information anytime anywhere. MobiVoD, the proposed solution, overcomes many difficulties currently challenging video streaming in a mobile ad hoc network. The new environment includes a three-tier architecture, in which the mobile VOD system employs a periodic broadcast protocol to achieve maximum scalability; and the clients leverage an ad hoc network caching technique to minimize the service delay. This system can sustain client failure and mobility, and provide true VOD services to most clients.

Given that the Internet does not widely support Internet protocol multicast while content-distribution-network technologies are costly, the concept of peer-to-peer could be a promising start for enabling large-scale streaming systems. In our so-called Zigzag approach, we propose a method for clustering peers into a hierarchy called the administrative organization for easy management, and a method for building the multicast tree atop this hierarchy for efficient content transmission. In Zigzag, the multicast tree has a height logarithmic with the number of clients, and a node degree bounded by a constant. This helps reduce the number of processing hops on the delivery path to a client while avoiding network bottlenecks. Consequently, the end-to-end delay is kept small. Although one could build a tree satisfying such properties easily, an efficient control protocol between the nodes must be in place to maintain the tree under the effects of network dynamics. Zigzag handles such situations gracefully, requiring a constant amortized worst-case control overhead. Especially, failure recovery is done regionally with impact on, at most, a constant number of existing clients and with mostly no burden on the server.

A peer-to-peer technique called ZIGZAG for single-source media streaming is designed . ZIGZAG allows the media server to distribute content to many clients by organizing them into an appropriate tree rooted at the server. This application-layer multicast tree has a height logarithmic with the number of clients and a node degree bounded by a constant. This helps reduce the number of processing hops on the delivery path to a client while avoiding network bottleneck. Consequently, the end-to-end delay is kept small. Although one could build a tree satisfying such properties easily, an efficient control protocol between the nodes must be in place to maintain the tree under the effects of network dynamics and unpredictable client behaviors. ZIGZAG handles such situations gracefully requiring a constant amortized control overhead. Especially, failure recovery can be done regionally with little impact on the existing clients and mostly no burden on the server.

Despite the amelioration in communication technologies, the server network I/O bottleneck remains a severe obstacle in video-on-demand applications. All bandwidth-sharing techniques, proposed to overcome this, assume videos to be requested in their entirety. In this paper, we extend the video-on-demand delivery problem so that it allows users to request parts of video and be able to receive them at that instant. Specifically, we propose an efficient technique called multicast jumping (MJ) that provides many desirable properties such as: (1) the demand on the server bandwidth is significantly alleviated; (2) true and scaleable video on demand is achieved; (3) not only for streaming video segments on demand, but the technique also works for on-demand videos in their entirety.

Focuses on the problem of providing quality-of-service guarantee, scalability, and on-demand property to video streaming systems. We propose a solution called layered range multicast (LRM). LRM allows transmitting a range of data to a multicast group's members, which helps clients who request for service at different times join a same multicast efficiently without additional server bandwidth allocation. This is more advanced than the conventional multicast in which a late client joining an existing multicast, without bandwidth support from the server, must miss a certain portion of the video requested. Another advantage of LRM is its capability to support clients requesting various levels of service quality. In addition, LRM does not assume the existence of IP Multicast and therefore it can be implemented on the current Internet without degrading the very inefficient centralized approach. Our performance study results confirm the above benefits.

We design a peer-to-peer technique for single-source media streaming. This technique allows the media server to distribute content to many clients by organizing them into an appropriate tree rooted at the server. This application-layer multicast tree has a height O(logN) where N is the number of clients, and a node degree bounded by a constant. This helps reduce the number of processing hops on the delivery path to a client while avoiding network bottleneck. Consequently, the end-to-end delay is kept small. Although one could build a tree satisfying such properties easily, an efficient control protocol between the nodes must be in place to maintain the tree under the effects of network dynamics and unpredictable client behaviors. Our technique handles such situations gracefully requiring a constant amortized control overhead. Especially, failure recovery can be done regionally with little impact on the existing clients.

In limited and asymmetric bandwidth environments such as wireless networks, push strategy can be used to allow a large number of mobile users to access the shared data. Most of today's solutions assume that the server disseminates data on only one physical channel. In this paper, we focus on the problem of broadcasting data over multiple physical channels that cannot be coalesced into fewer high-bandwidth ones. This introduces new hurdles such as the heterogeneities of channels and data set. We propose a novel broadcast design at the server side, which is aimed at minimizing the response and download times. We also present a caching policy at the client side to further enhance the performance. Theoretical foundations and simulation-based studies are also presented to substantiate the near-optimality and to assess the advantages of the proposed technique.