Recently, many graph based hashing methods have been emerged to tackle large-scale problems. However, there exists two major bottlenecks: (1) directly learning discrete codes is an NP-hardoptimization problem; (2) the complexity of both storage and computational time build a with n data points O(n2). To address these problems, in this paper, we propose novel yetsimple supervised method, asymmetric hashing, by preserving constraint building affinity matrix learn compact binary...

We study the cooperative asynchronous multi-agent multi-armed bandits problem, where each agent's active (arm pulling) decision rounds are asynchronous. That is, in round, only a subset of agents is to pull arms, and this unknown time-varying. consider two models cooperation, fully distributed leader-coordinated, propose algorithms for both that attain near-optimal regret communications bounds, which almost as good their synchronous counterparts. The algorithm relies on novel communication...

The online knapsack problem is a classic resource allocation in networking and operations research. Its basic version studies how to pack arriving items of different sizes values into capacity-limited knapsack. In this paper, we study general that includes item departures, while also considering multiple knapsacks multi-dimensional sizes. We design threshold-based algorithm prove the can achieve order-optimal competitive ratios. Beyond worst-case performance guarantees, aim near-optimal...

In many online learning paradigms, convexity plays a central role in the derivation and analysis of algorithms. The results, however, fail to be extended non-convex settings, which are necessitated by tons recent applications. Online Non-Convex Learning problem generalizes classic Convex Optimization framework relaxing assumption on cost function (to Lipschitz continuous function) decision set. state-of-the-art result for ønco demonstrates that Hedge algorithm attains sublinear regret O(√T...

In many online learning paradigms, convexity plays a central role in the derivation and analysis of algorithms. The results, however, fail to be extended non-convex settings, while non-convexity is necessitated by large number recent applications. Online Non-Convex Learning (ønco) problem generalizes classic Convex Optimization (øco) framework relaxing assumption on cost function (to Lipschitz continuous function) decision set. state-of-the-art result for ønco demonstrates that exponential...

The QoS (Quality of Service) buffer management problem, with significant and diverse computer applications, e.g., in online cloud resource allocation problems, is a classic admission control problem the presence constraints. In its basic setting, packets different values according to their requirements, arrive fashion switching node limited size. Then, switch needs make an immediate decision either admit or reject incoming packet based on value availability. objective maximize cumulative...

This paper proposes online offering strategies for a storage-assisted renewable power producer that participates in hour-ahead electricity market. The strategy determines the price and volume, while no exact or stochastic future information is available time-coupled setting presence of storage. proposed achieves best possible competitive ratio O(log θ), where θ between maximum minimum clearing prices. Trace-driven experiments demonstrate close-to-optimal performance.

This paper considers the problem of online linear optimization with inventory management constraints. Specifically, we consider an scenario where a decision maker needs to satisfy her time-varying demand for some units asset, either from market price or own inventory. In each time slot, is presented (linear) and must immediately decide amount purchase covering and/or storing in future use. The has limited capacity can be used buy store assets at low cover when high. ultimate goal slot while...

In this paper, we study the online multidimensional knapsack problem (called OMdKP) in which there is a whose capacity represented m dimensions, each dimension could have different capacity. Then, n items with scalar profit values and m-dimensional weights arrive an manner goal to admit or decline upon their arrival such that total obtained by admitted maximized of across all dimensions respected. This natural generalization classic single-dimension finds several relevant applications as...

This paper tackles a multi-agent bandit setting where M agents cooperate together to solve the same instance of K-armed stochastic problem. The are heterogeneous: each agent has limited access local subset arms and asynchronous with different gaps between decision-making rounds. goal for is find its optimal arm, can by sharing their observations others. While cooperation improves performance learning, it comes an additional complexity communication agents. For this heterogeneous setting, we...

This paper considers the problem of online linear optimization with inventory management constraints. Specifically, we consider an scenario where a decision maker needs to satisfy her timevarying demand for some units asset, either from market time-varying price or own inventory. In each time slot, is presented (linear) and must immediately decide amount purchase covering and/or storing in future use. The has limited capacity can be used buy store assets at low cover when high. ultimate goal...

Geographical load balancing takes advantage of the regional differences in dynamic electricity rates by shifting computing tasks among geographically distributed data centers. Since energy storage is becoming an integral part centers, one can maximize benefit temporal and spatial fluctuations combining geographical management. Previously, problem integrated with has been studied based on Lyapunov stochastic optimization approach, which relies asymptotic analysis averaging over infinite time...

The QoS buffer management problem, with significant and diverse computer applications, e.g., in online cloud resource allocation problems, is a classic admission control problem the presence of constraints. In its basic setting, packets different values, arrive fashion to switching node limited size. Then, switch needs make an immediate decision either admit or reject incoming packet based on value availability. objective maximize cumulative profit admitted packets, while respecting...

Although Adaptive RED (ARED) algorithms, which can adjust RED's parameters adaptively are broadly researched to deal with dynamic network scenarios, including bursty traffic and varying link state, unstability is still a constraint against the improvement of efficiency. In this paper, TCP/RED system considered as feedback where equilibrium exists. Through analyzing proposed system, adverse effects busty state studied respectively. We also point out that, exist ARED algorithms low-efficiency...

In this study, the authors develop a novel approach to class of pursuit‐evasion problems modelled in form discrete time feedback control systems, where opposing parties have asymmetric capability. The assume policy evader, described as random variable, is unknown pursuer. This problem formulated quadratic optimisation from perspective Due curse dimensionality, cannot be practically solved by dynamic programming. reformulate it multi‐armed bandit problem. A heuristic based on Gittins index...

The ever-increasing user requests for video services have posed a great challenge to the cellular networks, and emergence of various new also puts forward higher requirements mobile networks. By caching contents in cache-enabled heterogeneous delivery delay content stress backhaul links can be improved conspicuously. However, how store diverse has got much attentions past decade. In this paper, considering time-varying requests, hierarchical cooperative strategy with preference is proposed....

Motivated by the application of energy storage management in electricity markets, this paper considers problem online linear programming with inventory constraints. Specifically, a decision maker should satisfy some units an asset as her demand, either form market time-varying price or from own inventory. The is presented slot-by-slot manner, and must immediately decide purchased amount current to cover demand store for covering future demand. has limited capacity its critical role buy...

With the rapid growth of medical and biomedical image data, energy-efficient solutions for analyzing such data that can be processed fast accurately on platforms with low power budget are highly desirable. This paper uses segmenting glial cells in brain microscopy images as a case study to demonstrate how achieve segmentation significant energy saving minimal comprise accuracy. Specifically, we design, train, implement, evaluate Fully Convolutional Networks (FCNs) IBM's neurosynaptic DNN...