We study the problem of learning to rank from multiple information sources. Though multi-view learning and learning to rank have been studied extensively leading to a wide range of applications, multi-view learning to rank as a synergy of both topics has received little attention. The aim of the paper is to propose a composite ranking method while keeping a close correlation with the individual rankings simultaneously. We present a generic framework for multi-view subspace learning to rank (MvSL2R), and two novel solutions are introduced under the framework. The first solution captures information of feature mappings from within each view as well as across views using autoencoder-like networks. Novel feature embedding methods are formulated in the optimization of multi-view unsupervised and discriminant autoencoders. Moreover, we introduce an end-to-end solution to learning towards both the joint ranking objective and the individual rankings. The proposed solution enhances the joint ranking with minimum view-specific ranking loss, so that it can achieve the maximum global view agreements in a single optimization process. The proposed method is evaluated on three different ranking problems, i.e., university ranking, multi-view lingual text ranking, and image data ranking, providing superior results compared to related methods.

In this paper, we proposed a novel deep-learning method called Inception LSTM for video frame prediction. A standard convolutional LSTM uses a single size kernel for each of its gates. Having multiple kernel sizes within a single gate would provide a richer features that would otherwise not be possible with a single kernel. Our key idea is to introduce inception like kernels within the LSTM gates to capture features from a bigger area of the image while retaining the fine resolution of small information. We implemented the proposed idea of inception LSTM network on PredNet network with both inception version 1 and inception version 2 modules. The proposed idea was evaluated on both KITTI and KTH data. Our results show that the Inception LSTM has better predictive performance compared to convolutional LSTM. We also observe that LSTM with Inception version 1 has better predictive performance compared to Inception version 2, but Inception version 2 has less computational cost.

Link prediction refers to estimating the likelihood of a link appearing in the future based on the current status of a graph. Link prediction problem applications in various domains such as bioinformatics, social network analysis, cybersecurity and e-commerce. Some of these graphs are massive and are constantly evolving. Many applications require these graph streams to be processed them in real-time, to predict the link based on the most recent information as the graph features may change over time. Existing approaches to process large graphs for link prediction is non-trivial due to the following reasons: 1) Graphs required to predict the links are too large to be stored in a single RAM. Link prediction on these large graphs is expensive in terms of computation resources and time required to perform link prediction, 2) Sketch-based approaches are not suitable in applications where accuracy is critical (such as analyzing criminal social networks or supply chain networks) and 3) Sketch-based approaches also fail to handle dynamic graphs, where edges are not only added, but also removed. This results in changes to the graph topology, making the features previously computed to be obsolete. Distributed data stream frameworks such as Apache Flink could be potentially used for distributed graph processing. However, there are no techniques to handle link prediction on distributed graph streams. In this paper, we consider three fundamental, neighborhood-based link prediction measures, Jaccard coefficient, Preferential attachment, and common neighbors and enable an accurate measurement of them to address link prediction problem in dynamic graph streams. We propose a neighborhood-centric graph processing approach to handle graphs that exploits the locality, parallelism, and incremental computation of existing distributed frameworks to calculate these graph features with exact results. We perform experimental studies on various real-world graph streams. The results demonstrate that our graph measures are accurate and are more efficient than the existing vertex-centric approaches to graph processing.

Solar irradiance is the measurement of the amount of power from the Sun per unit area. Solar irradiance has a very high degree of variability, due to many environmental factors, including cloud cover, relative humidity, and air temperature. Predicting solar irradiance is very useful for measuring future solar energy production and power scheduling. Real-time solar irradiance can be forecasted using either maching learning or physics-based models, both having their own respective trade-offs. To overcome the limitations of both machine learning and physics-based simulations, we propose a novel framework in predictive analytics that combines the power of both types of models. In particular, we propose a concept-drift framework to develop a model that integrates satellite cloud imagery with real-time solar irradiance measurements to improve irradiance forecasting. We present preliminary results using real-world irradiance and cloud data to support the motivation for a concept-drift based approach. Based on our results, we found that one static model cannot fully capture the dynamics of solar irradiance on both days where cloud cover is present and days where there is no cloud cover.

This paper proposes a novel unsupervised rank aggregation method using parameterized function optimization (PFO). This algorithm derives a parameterized rank aggregation model by minimizing the energy of weighted standard deviations of rank lists associated with different rankers or attributes. Parameters, in this problem, are weights representing the impact of rank lists on the final aggregated rank. The proposed learning to rank aggregation method is efficient (linear time complexity) and its accuracy compares favorably with pairwise preference methods (with polynomial time complexity). Two rounds of experiments are run to show the success of PFO in rank aggregation: one on the learning to rank (LETOR) benchmark dataset to show its success in unsupervised rank aggregation and the other on three university ranking datasets to solve a practical problem in education. The experimental results on the LETOR show that PFO significantly outperforms the baseline results and show promising performances in comparison with recent high performance methods developed for unsupervised rank aggregation. The university ranks obtained by our model compare favorably with the ranks reported by well-known organizations. Success of the PFO model for performing unsupervised rank aggregation, specifically on practical problems, supports the use of the algorithm in difficult ranking scenarios without ground truth.

Video traffic accounts for more than 70% of Internet traffic and is likely to reach 90% in 2020. Real-time video streaming is used in many applications such as sports, video conferencing, video surveillance, and public safety. Video streaming protocols such as MPEG-DASH, HDS and RTMP provide adaptive streaming capability, and application-level fault tolerance to handle failures while providing uninterrupted live video streaming service. Network outages and congestion is common in many situations during emergencies, or large public events. These failures impact live video streaming performance in terms of packet loss, latency, jitter, and video quality. Both transport layer and video streaming protocols have in-built fault tolerance mechanisms. Integrating streaming protocol's fault handling mechanism with the automatic network fault recovery would provide optimized video streaming performance. The goal of this paper is to investigate how different video streaming protocols perform when fault tolerance is provided by the SDN network. We created a self-healing SDN based MESH network testbed that transports video from a live source to the server. Our experimental results show that RTMP outperforms both MPEG- DASH, and HDS in terms of live video stream recovery and resiliency to link, port, switch, and or controller failures. These results were obtained under controlled network conditions with acceptable latency and jitter. However, switching time for recovery is significantly higher than traditional network setup, but with SDN better programable and automated recovery is possible than traditional network.

For the last decade, the automatic generation of hypothesis from the literature has been widely studied. One common approach is to model biomedical literature as a concept network; then a prediction model is applied to predict the future relationships (links) between pairs of concept. Typically, this link prediction task can be cast into in one of two forms: (a) predict the future links for a specific concept (node) or (b) predict the future links for the entire network. However, while being able to accurately forecast future relationships is vital, another, equally important question should be addressed: of the predicted links, which will be most important and/or most relevant? Attempts to answer these questions in the past have generally been domain specific. In this paper, we propose a domain-independent, supervised method that predicts the rank of future links utilizing objective interestingness measures. The results, based on analysis of thirteen common interestingness measures, indicate that, while predicting the specific future interestingness values is difficult, our approach allowed us to capture the relative ordering of the links with low error.

Reliability estimation of High Performance Computing (HPC) systems enables resource allocation, and fault tolerance frameworks to minimize the performance loss due to unexpected failures. Recent studies have shown that compute nodes in HPC systems follow a time varying failure rate distribution such as Weibull, instead of the exponential distribution. In this paper, we propose a model for the Time to Failure (TTF) distribution of a system of k s-independent nodes when individual nodes exhibit time varying failure rates. We also present the system reliability, failure rates, Mean Time to Failure (MTTF), and derivations of the proposed system TTF model. The model is validated using observed data on time to failure.

Failures and downtimes have severe impact on the performance of parallel programs in a large scale High Performance Computing (HPC) environment. There were several research efforts to understand the failure behavior of computing systems. However, the presence of multitude of hardware and software components required for uninterrupted operation of parallel programs make failure and reliability prediction a challenging problem. HPC run-time systems like checkpoint frameworks and resource managers rely on the reliability knowledge of resources to minimize the performance loss due to unexpected failures. In this paper, we first analyze the Time Between Failure (TBF) distribution of individual nodes from a 512-node HPC system. Time varying distributions like Weibull, lognormal and gamma are observed to have better goodness-of-fit as compared to exponential distribution. We then present a reliability-aware resource allocation model for parallel programs based on one of the time varying distributions and present reliability-aware resource allocation algorithms to minimize the performance loss due to failures. We show the effectiveness of reliability-aware resource allocation algorithms based on the actual failure logs of the 512 node system and parallel workloads obtained from LANL and SDSC. The simulation results indicate that applying reliability-aware resource allocation techniques reduce the overall waste time of parallel jobs by as much as 30%. A further improvement by 15% in waste time is possible by considering the job run lengths in reliability-aware scheduling.