Rankings and scores are two common data types used by judges to express preferences and/or perceptions of quality in a collection of objects. Numerous models exist to study data of each type separately, but no unified statistical model captures both data types simultaneously without first performing data conversion. We propose the Mallows-Binomial model to close this gap, which combines a Mallows $\phi$ ranking model with Binomial score models through shared parameters that quantify object quality, a consensus ranking, and the level of consensus among judges. We propose an efficient tree-search algorithm to calculate the exact MLE of model parameters, study statistical properties of the model both analytically and through simulation, and apply our model to real data from an instance of grant panel review that collected both scores and partial rankings. Furthermore, we demonstrate how model outputs can be used to rank objects with confidence. The proposed model is shown to sensibly combine information from both scores and rankings to quantify object quality and measure consensus with appropriate levels of statistical uncertainty.

Conventional variational autoencoders fail in modeling correlations between data points due to their use of factorized priors. Amortized Gaussian process inference through GP-VAEs has led to significant improvements in this regard, but is still inhibited by the intrinsic complexity of exact GP inference. We improve the scalability of these methods through principled sparse inference approaches. We propose a new scalable GP-VAE model that outperforms existing approaches in terms of runtime and memory footprint, is easy to implement, and allows for joint end-to-end optimization of all components.

Bayesian optimization has recently emerged as a popular method for the sample-efficient optimization of expensive black-box functions. However, the application to high-dimensional problems with several thousand observations remains challenging, and on difficult problems Bayesian optimization is often not competitive with other paradigms. In this paper we take the view that this is due to the implicit homogeneity of the global probabilistic models and an overemphasized exploration that results from global acquisition. This motivates the design of a local probabilistic approach for global optimization of large-scale high-dimensional problems. We propose the TuRBO algorithm that fits a collection of local models and performs a principled global allocation of samples across these models via an implicit bandit approach. A comprehensive evaluation demonstrates that TuRBO outperforms state-of-the-art methods from machine learning and operations research on problems spanning reinforcement learning, robotics, and the natural sciences.

Some errors were introduced while preparing the final source files for the original article published in August 2013 in the 9,4 issue of TOMM. The errata are summarized here below together with attached revised pages showing the corrected elements indicated in red. The full CVoR (Corrected Version of Record) can be accessed in the ACM Digital Library, DOI=http://dx.doi.org/10.1145/2501643.2501652 —Page 8: New Figure 6(a) —Page 16: New Figures 8(a), 8(b), and 9(a) —Page 17: New Figure 10(b) —Page 18: New Figures 11 and 12; corrected text reference —Page 19: Final sentence deleted

Video on Demand (VoD) services allow users to select and locally consume remotely stored content. We investigate the use of caching to solve the scalability issues of several existing VoD providers. We propose metrics and goals that define the requirements of a caching framework for CDNs of VoD systems. Using data logs collected from Motorola equipment from Comcast VoD deployments we show that several classic caching solutions do not satisfy the proposed goals. We address this issue by developing novel techniques for predicting future values of several metrics of interest. We rely on computed predictions to define the penalty imposed on the system, both network and caching sites, when not storing individual items. We use item penalties to devise novel caching and static content placement strategies. We use the previously mentioned data logs to validate our solutions and show that they satisfy all the defined goals.

Video on Demand (VoD) services allow users to select and locally consume remotely stored content. We investigate the use of caching to solve the scalability issues of several existing VoD providers. We propose metrics and goals that define the requirements of a caching framework for CDNs of VoD systems. Using data logs collected from Motorola equipment from Comcast VoD deployments we show that several classic caching solutions do not satisfy the proposed goals. We address this issue by developing a novel technique for predicting future values of several metrics of interest. We use these predictions to evaluate the penalty imposed on the system (network and caches) when not caching individual items. We use item penalties to propose novel caching and static placement strategies. We use the mentioned data logs to validate our solutions and show that they satisfy all the defined goals.

Video on Demand (VoD) services provide a wide range of content options and enable subscribers to select, retrieve and locally consume desired content. In this work we propose caching solutions to improve the scalability of the content distribution networks (CDNs) of existing VoD architectures. We first investigate metrics relevant to this caching framework and subsequently define goals that should be satisfied by an efficient solution. We propose novel techniques for predicting future values of metrics of interest. We use our prediction mechanisms to define the cost imposed on the system (network and caches) by items that are not cached. We use this cost to develop novel caching and static placement strategies. We validate our solutions using log data collected from Motorola equipment from several Comcast VoD deployments.

Purpose The purpose of this paper is to address some weaknesses in the handling of current multi-factor authentication, suggests some criteria for overcoming these weaknesses and presents a simple proof of concept authentication system. Design/methodology/approach First, this paper evaluates some of the underlying practices and assumptions in multi-factor authentication systems. Next, the paper assesses the implications of these when compared to a quantitative authentication risk management approach. Based upon these implications this paper next note the requirements for an improved system and detail some related research areas that meet these requirements. Finally, this paper discussed how a system that meets these requirements through the application of that research could provide benefits and outlined a simple points-based authentication system. Findings The paper proposes that many of the weaknesses in authentication confidence management could be effectively mitigated through the deployment of a factor independent multi-modal fusion quantitative authentication-based system. This paper details a simple point-based approach that does this and discuss how addressing the problems in handling authentication confidence could further optimise risk management in multi-factor authentication systems. Practical implications This paper's suggestions for optimising multi-factor authentication have many implications within medium to high-security commercial and government applications. Correct authentication risk handling enables decisions regarding risk and authentication to be made more accurately. Originality/value This implications of the issues discussed in this paper have relevance to anyone who deploys or uses any medium to high-security authentication system. As the bottom end of the medium to high-security range includes online banking, there are implications for a wide range of stakeholders.

Existing mechanisms for querying wireless sensor networks leak client interests to the servers performing the queries. The leaks are not only in terms of specific regions of interest but also of client access patterns. In this article we introduce the problem of preserving the privacy of clients querying a wireless sensor network owned by untrusted organizations. We first propose an efficient protocol, SPYC, that ensures full client privacy in settings where the servers providing access to the network are honest-but-curious and whose collaboration does not extend beyond well-defined administrative purposes. Furthermore, we study the same query privacy problem in a setting where servers exhibit malicious behavior or where powerful external attackers have access to sensor network traffic information. In this setting we propose two metrics for quantifying the privacy achieved by a client's query sequence. We then extend SPYC with a suite of practical algorithms, then analyze the privacy and efficiency levels they provide. Our TOSSIM simulations show that the proposed extensions are communication efficient while significantly improving client privacy levels.

We present techniques for synchronizing nodes that periodically broadcast content and presence updates to colocated nodes over an ad hoc network, where nodes may exhibit Byzantine malicious behavior. Instead of aligning duty cycles, our algorithms synchronize the periodic transmissions of nodes. This allows nodes to save battery power by switching off their network cards without missing updates from their neighbors. We propose several novel attack classes and show that they are able to disrupt synchronization even when launched by a single attacker. Finally, we devise a rating based algorithm (RBA) that rates neighbors based on the consistency of their behavior. By favoring well-behaved nodes in the synchronization process, we show that RBA quickly stabilizes the synchronization process and reduces the number of lost updates by 85 percent. Our evaluation also shows that all our algorithms are computationally efficient and, for the setup considered, extend the device lifetime by 30 percent over an always-on Wi-Fi scenario.

Richer communication, computational and content- handling capabilities on mobile devices are driving the adoption of more interactive and collaborative applications. These motivate a need for data-sharing paradigms that are flexible, efficient and lightweight - but also accessible across the different applications that may be involved in an ongoing user task. Our AppBus framework provides a unique "short term memory" facility for mobile devices, facilitating such data-sharing across co-resident applications. With the Remote AppBus extension, we make this facility accessible in a consistent manner across address spaces on a device, and potentially across devices. In this paper, we describe the design and implementation of remote AppBus, and discuss its utility in enabling task collaboration on device, and in facilitating task continuity across devices. We conclude with a look at related efforts, challenges and future directions for the research.

In this paper, we present techniques for synchronizing nodes that periodically broadcast content and presence updates to co-located nodes over an ad-hoc network, where nodes may exhibit Byzantine malicious behavior. We first propose an algorithm for synchronizing the periodic transmissions of all the nodes in an attacker-free multi-hop network. This allows nodes to save battery power by switching off their network cards without missing updates from their neighbors. We then introduce a suite of spoofing attacks and show that they are able to disrupt synchronization and destabilize the network even when launched by a single attacker in large, multi-hop networks. Finally, we devise a rating based algorithm that rates neighbors based on the consistency of their behaviors. By favoring well-behaved nodes in the synchronization process, we show that we can address the issue of Byzantine malicious behavior very effectively. Our evaluation shows that the algorithms are computationally efficient and, for the setup considered, extend the device lifetime by 30% over an always-on Wi-Fi scenario. Moreover, in the presence of attacks, our rating based algorithm quickly stabilizes the synchronization process and reduces the number of lost updates by 85%.

Even as mobile devices as well as wireless networks become increasingly capable, newer content-intensive application place increasing demands on the resources of these devices. Some of these applications make conflicting resource demands making it difficult for device platforms to guarantee acceptable user experience. Thus, a resource aware middleware is needed that mobile applications can leverage to manage their content. In this paper we discuss Cache & Carry, a framework for mobile devices that intelligently manages scarce resources on the mobile device across multiple applications. The framework takes into consideration changing mobile device characteristics such as available battery and storage as well as changing network characteristics such as availability of multiple networks, their perceived cost, their available bandwidths etc., to provide an optimal end user experience.

Encouraging cooperation between users of mobile devices operating in ad hoc mode is a difficult task mostly because the scarce battery and bandwidth resources of devices suggest that selfish behavior may be most beneficial. The insecure usage of credits to reward cooperation can easily provide an incentive for cheating, thus, on the long term only leading to selfishness. In this paper we propose several secure credit based mechanisms enforcing fairness in a hybrid wireless content retrieval system operating both in cellular and ad hoc connectivity modes. Our solution consists of mechanisms for securely and privately discovering desired content on neighboring devices, simultaneously exchanging credit and content shares in a verifiable manner and for generating and expiring non-forgeable credits. We present experimental results of a partial prototype of our system implemented on MPx and E680i cellular phones and HP iPaq hx4700 PDAs, along with extensive simulation results showing that our solution significantly reduces the effectiveness of selfish behavior, making it an unattractive strategy.

The increasing quality of computation and connectivity on mobile devices has motivated a need for data-sharing between resident services. Most such services rely on the user to transfer data manually between them. This approach is not only prone to input error but also creates a significant data-entry burden for the user on a mobile device. With AppBus, we envision a cooperative data-sharing facility that provides the equivalent of a “short-term memory” for the mobile device. By leveraging this transient data store, services can collaborate in a loosely-coupled manner that makes their combined operation appear more intuitive and seamless. In this paper, we describe the design and implementation of AppBus for a mobile phone and initiate a discussion on both the utility and the limitations of this paradigm in mobile computing.

This paper explores the application of genetic algorithms to the learning of local robot navigation behaviors for reactive control systems. Our approach is to train a reactive control system in various types of environments, thus creating a set of "ecological niches" that can be used in similar environments. The use of genetic algorithms as an unsupervised learning method for a reactive control architecture greatly reduces the effort required to configure a navigation system. Findings from computer simulations of robot navigation through various types of environments are presented. I. Introduction A common robot task is to navigate through an environment to a goal position, without hitting any obstacles that may be present. Navigation through a cluttered environment is an extremely complex and underconstrained task. Apart from the computational constraints placed on the design of a navigation system, it is desirable that the system be robust enough to navigate through a large number ...