Existing constrained multiobjective evolutionary algorithms (CMOEAs) frequently employ the information provided by the unconstrained Pareto front (UPF) to facilitate the identification of the constrained Pareto front (CPF) for constrained multiobjective optimization problems (CMOPs). However, obtaining a UPF with favorable convergence and diversity is not straightforward, and the obtained UPF is sometimes difficult to effectively assist in the identification of CPF for certain complex CMOPs. To this end, a novel algorithm called DREMCO is proposed, which endeavors to obtain a good UPF and is capable of utilizing the obtained UPF to consistently assist in the identification of CPF. DREMCO consists of a main population for the original problem and a two-phase (propulsion phase and recovery phase) auxiliary population with a dynamics regulation mechanism. In the propulsion phase, the auxiliary population ignores constraints and employs an improved aggregation function to obtain a good UPF, thereby pulling the main population across infeasible regions. In the recovery phase, the auxiliary population uses a penalty function method to converge to CPF and continuously refines the CPF of the main population. Concurrently, a novel phase judgment method is proposed for seamless transition between phases. Furthermore, an information-sharing strategy is proposed, which is capable of sharing information of the parents and offspring in offspring generation and environment selection, respectively. The experimental results with 11 state-of-the-art CMOEAs on five benchmark suites and eight real-world CMOPs demonstrate the efficacy of the proposed algorithm.

On-device models directly expose model details such as parameters and architectures to threats, enabling model theft and white-box attacks. Existing solutions face trade-offs: encryption struggle with the trade-off between effectiveness and efficiency; obfuscation introduces significant runtime overhead and cannot support unseen obfuscated models; hardware-based approaches require specific hardware support that can hardly be used by third-party apps. We propose ModelVirtualizer, virtualizing metadata at four semantic domains using symmetric encryption algorithms (e.g., AES-256-CTR), performing lightweight decryption during loading. In addition, we introduce modular arithmetic in encryption-decryption to eliminate the connection between model details (operator types, graph connections), which allows the virtual machine to support unseen virtualized models and resist reverse engineering. Evaluation on $\mathbf{1 0}$ models shows functional equivalence between virtualized models and natives, with comparable loading and inference performance to native TFLite, and performance advantages over obfuscation-based approaches.

Existing Graph Neural Network (GNN) training frameworks have been designed to help developers easily create performant GNN implementations. However, most existing GNN frameworks assume that the input graphs are static, but ignore that most real-world graphs are constantly evolving. Though many dynamic GNN models have emerged to learn from evolving graphs, the training process of these dynamic GNNs is dramatically different from traditional GNNs in that it captures both the spatial and temporal dependencies of graph updates. This poses new challenges for designing dynamic GNN training frameworks. First, the traditional batched training method fails to capture real-time structural evolution information. Second, the time-dependent nature makes parallel training hard to design. Third, it lacks system supports for users to efficiently implement dynamic GNNs. In this paper, we present NeutronStream, a framework for training dynamic GNN models. NeutronStream abstracts the input dynamic graph into a chronologically updated stream of events and processes the stream with an optimized sliding window to incrementally capture the spatial-temporal dependencies of events. Furthermore, NeutronStream provides a parallel execution engine to tackle the sequential event processing challenge to achieve high performance. NeutronStream also integrates a built-in graph storage structure that supports dynamic updates and provides a set of easy-to-use APIs that allow users to express their dynamic GNNs. Our experimental results demonstrate that, compared to state-of-the-art dynamic GNN implementations, NeutronStream achieves speedups ranging from 1.48X to 5.87X and an average accuracy improvement of 3.97%.

Key-based workload partitioning is a common strategy used in parallel stream processing engines, enabling effective key-value tuple distribution over worker threads in a logical operator. It is likely to generate poor balancing performance when workload variance occurs on the incoming data stream. This paper presents a new key-based workload partitioning framework, with practical algorithms to support dynamic workload assignment for stateful operators. The framework combines hash-based and explicit key-based routing strategies for workload distribution, which specifies the destination worker threads for a handful of keys and assigns the other keys with the hash function. When short-term distribution fluctuations occur to the incoming data stream, the system adaptively updates the routing table containing the chosen keys, in order to rebalance the workload with minimal migration overhead within the stateful operator. We formulate the rebalance operation as an optimization problem, with multiple objectives on minimizing state migration costs, controlling the size of the routing table and breaking workload imbalance among worker threads. Despite of the NP-hardness nature behind the optimization formulation, we carefully investigate and justify the heuristics behind key (re)routing and state migration, to facilitate fast response to workload variance with ignorable cost to the normal processing in the distributed system. Empirical studies on synthetic data and real-world stream applications validate the usefulness of our proposals.

Proper admission control in cognitive radio networks is critical in providing QoS guarantees to secondary unlicensed users. In this paper, we study the admission control and channel allocation problem in overlay cognitive radio networks under the maximum cumulative delay constraint. We formulate it as a Markov decision process problem, and then solve it by transforming the original formulation into a stochastic shortest path problem. We further simulate the performance of a class of threshold-based admission control with the largest-delay-first channel allocation policy, and show its advantage over other two benchmark policies.

Several studies have shown that in high-speed networks it is likely that a large number of packets are dropped in a single loss event on a bottleneck link and a large portion of the ongoing flows are about to experience packet losses. Such synchronized losses among flows have negative effects on the performance of most loss-based congestion control protocols proposed for high-speed networks, especially with respect to fairness considerations. We propose in this paper F2-TCP, a new loss-based congestion control protocol for high-speed networks, to achieve better fairness and TCP-friendliness. Simulation studies show that F2-TCP indeed fulfills these design goals. The superior performance and the low deployment cost, especially on legacy machines, make F2-TCP a good alternative for todaypsilas high-speed networks.

Many sensor network applications only require probabilistic data delivery, as they can tolerate some missing data samples. For example, in environmental monitoring, missing temperature, pressure and humidity level samples can often be inferred by spatial and/or temporal interpolations. In this paper we propose and study an adaptive p-persistent CSMA-based media access control protocol that supports end-to-end probabilistic reliability for sensor networks on a hop-by-hop basis. In an effort to reduce the probability of packet collisions, first we tune the carrier sensing range of the nodes; then given an end-to-end reliability requirement, we determine the optimal allocation of per-hop reliability requirements on each route to minimize the expected total number of transmissions needed; finally, our adaptive p-persistent CSMA protocol tunes its link persistence probability to further reduce the expected total number of transmissions, and thereby minimizes the energy consumption in the network. We formulate this latter problem as a constrained optimization problem, and then derive an algorithm to adapt the link persistence probabilities using the Lagrangian dual decomposition method.

This paper studies the tradeoff between energy consumption and application performance in wireless sensor networks by investigating the interaction between network lifetime maximization and rate allocation problems. To guarantee the individual performance of sensor nodes, we adopt the network utility maximization (NUM) framework to ensure certain fairness on source rates of sensor nodes. We formulate the network lifetime maximization problem and fair rate allocation problem as constrained maximization problems, and combine them by introducing a system parameter, which characterizes the tradeoff between the two problems. Using Lagrange dual decomposition, the original problem is vertically decomposed into three subproblems: a rate control problem at the transport layer, a contention resolution problem at the MAC Layer, and a cross-layer energy conservation problem. The first and second subproblems jointly solve the congestion problem in sensor networks via congestion prices, and fully distributed algorithms are derived. Furthermore, they are coupled with the cross layer energy conservation problem to solve the network lifetime maximization problem via energy prices. For the third subproblem, we first propose a partially distributed algorithm where network lifetime is a global information, and then by exploring the similarity between max-min rate allocation and network lifetime maximization in sensor networks, we approximate the latter by the NUM framework, and hence formulate the tradeoff problem in the unified NUM framework. As a result, a fully distributed algorithm is derived for the energy conservation problem.

In this paper, we propose and study two adaptive media access control protocols to support probabilistic data delivery reliability in sensor networks using MAC layer retransmissions. Since retransmissions consume energy, our adaptive p-persistent CSMA protocols tune their persistence probabilities in an effort to minimize the expected number of retransmissions that meet a given probabilistic reliability requirement, and thereby minimize energy consumption in the network. We formulate this problem of reliability with minimal energy consumption at the MAC layer as a constrained optimization problem and then derive the algorithms to adapt the persistence probabilities using the Lagrangian dual decomposition method.

This paper considers the problem of routing in sensor networks from the point of view od data collection. That is, given the initial amount of battery energy in each node, the aim is to determine how much data can each source transmit until the network is partitioned (i. e., until the nodes cannot find end-to-end routes to their respective sinks). In addition, to respond to some specific applications' requirements, when determining such nodal data volume distribution, fairness among nodes is taken into account. The problem is formulated as a concave utility maximization and a sub-gradient algorithm is proposed to solve it distributively. Some numerical results are given and the convergence of the algorithm is discussed.

Sensor networks are deployed to gather some useful data from a field and forward it toward a set of base stations or sinks for data analysis and decision making. Each sensor node is endowed with a finite amount of energy, and each byte transmission or reception costs a certain fixed fraction of energy as well as a variable fraction that depends on the distance between sender and receiver. Maximizing the volume of data collected at the sinks until some particular set of nodes exhaust their battery and partition the network is a very desirable trait in sensor networks, as it equates with a high level of energy efficiency. In this paper we formulate the problem of maximum data collection routing for sensor networks as a utility maximization problem subject to energy constraints, and invoke lagrange relaxation, duality and sub-gradient technique to solve the problem. We then focus on the problem of path oscillation, which is well known to happen in routing algorithms where link costs are function of the traffic load and propose heuristic solutions to address this oscillation problem

The network lifetime and application performance are two fundamental but conflicting desig objectives in wireless sensor networks. Hence there is an intrinsic tradeoff between network lifetime maximization and application performance maximization. Often application performance correlates to the application data rate obtained in sensor networks. We can thus study this tradeoff by investigating the interactions between the network lifetime maximization problem and the rate allocation problem. Severe bias on the allocated rates of some sensor nodes may exist if only the total throughput of the sensor network is maximized, hence we enforce fairness on source rates of sensor nodes by invoking the network utility maximization (NUM) framework. First we consider the network lifetime as global information shared by sensor nodes. We formulate the network lifetime maximization and fair rate allocation both as constrained maximization problems. By introducing a system parameter, we combine these two objectives into a single weighted objective, and characterize the tradeoff between them. Then we give the optimality condition, and derive a partially distributed algorithm. Also, we identify the similarity between network lifetime maximization and max-min rate allocation in networks. Since the latter one can be approximated using NUM framework, we adopt the same idea for the former one, and approximate the optimal solution in the unified NUM framework. Based on this, an efficient fully distributed algorithm is derived.