Riemannian Stochastic Recursive Momentum Method for non-Convex Optimization

We propose a stochastic recursive momentum method for Riemannian non-convex optimization that achieves a nearly-optimal complexity to find epsilon-approximate solution with one sample. The new algorithm requires one-sample gradient evaluations per iteration and does not require restarting with a large batch gradient, which is commonly used to obtain a faster rate. Extensive experiment results demonstrate the superiority of the proposed algorithm. Extensions to nonsmooth and constrained optimization settings are also discussed.

Benchmarking Analysis of the Accuracy of Classification Methods Related to Entropy

In the machine learning literature we can find numerous methods to solve classification problems. We propose two new performance measures to analyze such methods. These measures are defined by using the concept of proportional reduction of classification error with respect to three benchmark classifiers, the random and two intuitive classifiers which are based on how a non-expert person could realize classification simply by applying a frequentist approach. We show that these three simple methods are closely related to different aspects of the entropy of the dataset. Therefore, these measures account somewhat for entropy in the dataset when evaluating the performance of classifiers. This allows us to measure the improvement in the classification results compared to simple methods, and at the same time how entropy affects classification capacity. To illustrate how these new performance measures can be used to analyze classifiers taking into account the entropy of the dataset, we carry out an intensive experiment in which we use the well-known J48 algorithm, and a UCI repository dataset on which we have previously selected a subset of the most relevant attributes. Then we carry out an extensive experiment in which we consider four heuristic classifiers, and 11 datasets.

Modeling the Stock Relation with Graph Network for Overnight Stock Movement Prediction

Stock movement prediction is a hot topic in the Fintech area. Previous works usually predict the price movement in a daily basis, although the market impact of news can be absorbed much shorter, and the exact time is hard to estimate. In this work, we propose a more practical objective to predict the overnight stock movement between the previous close price and the open price. As no trading operation occurs after market close, the market impact of overnight news will be reflected by the overnight movement. One big obstacle for such task is the lacking of data, in this work we collect and publish the overnight stock price movement dataset of Reuters Financial News. Another challenge is that the stocks in the market are not independent, which is omitted by previous works. To make use of the connection among stocks, we propose a LSTM Relational Graph Convolutional Network (LSTM-RGCN) model, which models the connection among stocks with their correlation matrix. Extensive experiment results show that our model outperforms the baseline models. Further analysis shows that the introduction of the graph enables our model to predict the movement of stocks that are not directly associated with news as well as the whole market, which is not available in most previous methods.

Accurate Lane Detection for Self-Driving Cars: An Approach Based on Color Filter Adjustment and K-Means Clustering Filter

Lane detection is a crucial factor for self-driving cars to achieve a fully autonomous mode. Due to its importance, lane detection has drawn wide attention in recent years for autonomous driving. One challenge for accurate lane detection is to deal with noise appearing in the input image, such as object shadows, brake marks, breaking lane lines. To address this challenge, we propose an effective road detection algorithm. We leverage the strength of color filters to find a rough localization of the lane marks and employ a K-means clustering filter to screen out the embedded noises. We use an extensive experiment to verify the effectiveness of our method. The result indicates that our approach is robust to process noises appearing in input image, which improves the accuracy in lane detection.

Packet Scheduling for Multiple-Switch Software-Defined Networking in Edge Computing Environment

Software-defined networking (SDN) decouples the control plane and data forwarding plane to overcome the limitations of traditional networking infrastructure. Among several communication protocols employed for SDN, OpenFlow is most widely used for the communication between the controller and switch. In this paper two packet scheduling schemes, FCFS-Pushout (FCFS-PO) and FCFS-Pushout-Priority (FCFS-PO-P), are proposed to effectively handle the overload issue of multiple-switch SDN targeting the edge computing environment. Analytical models on their operations are developed, and extensive experiment based on a testbed is carried out to evaluate the schemes. They reveal that both of them are better than the typical FCFS-Block (FCFS-BL) scheduling algorithm in terms of packet wait time. Furthermore, FCFS-PO-P is found to be more effective than FCFS-PO in the edge computing environment.

BGPcoin: Blockchain-Based Internet Number Resource Authority and BGP Security Solution

Without the design for inherent security, the Border Gateway Protocol (BGP) is vulnerable to prefix/subprefix hijacks and other attacks. Though many BGP security approaches have been proposed to prevent or detect such attacks, the unsatisfactory cost-effectiveness frustrates their deployment. In fact, the currently deployed BGP security infrastructure leaves the chance for potential centralized authority misconfiguration and abuse. It actually becomes the critical yield point that demands the logging and auditing of misbehaviors and attacks in BGP security deployments. We propose a blockchain-based Internet number resource authority and trustworthy management solution, named BGPcoin, to facilitate the transparency of BGP security. BGPcoin provides a reliable origin advertisement source for origin authentication by dispensing resource allocations and revocations compliantly against IP prefix hijacking. We perform and audit resource assignments on the tamper-resistant Ethereum blockchain by means of a set of smart contracts, which also interact as one to provide the trustworthy origin route examination for BGP. Compared with RPKI, BGPcoin yields significant benefits in securing origin advertisement and building a dependable infrastructure for the object repository. We demonstrate it through an Ethereum prototype implementation, and we deploy it and do experiment on a locally-simulated network and an official Ethereum test network respectively. The extensive experiment and evaluation demonstrate the incentives to deploy BGPcoin, and the enhanced security provided by BGPcoin is technically and economically feasible.

Deep View-Aware Metric Learning for Person Re-Identification

Person re-identification remains a challenging issue due to the dramatic changes in visual appearance caused by the variations in camera views, human pose, and background clutter. In this paper, we propose a deep view-aware metric learning (DVAML) model, where image pairs with similar and dissimilar views are projected into different feature subspaces, which can discover the intrinsic relevance between image pairs from different aspects. Additionally, we employ multiple metrics to jointly learn feature subspaces on which the relevance between image pairs are explicitly captured and thus greatly promoting the retrieval accuracy. Extensive experiment results on datasets CUHK01, CUHK03, and PRID2011 demonstrate the superiority of our method compared with state-of-the-art approaches.

What network topology can tell in election prediction

With the advent of big data era, it is more and more acceptable that the topology of a network can reveal more than we can conceive. In this paper, we propose an algorithm to predict a winner of an election among several competitors based on the relationships of individuals in a social network. Convergence is proved and an extensive experiment is done to show the effectiveness of the algorithm. Our algorithm can also be used to identify members of different parties in a fairly reasonable manner. Furthermore, our algorithm is merely based on the topology of the network, which saves us a large amount of work from collecting voting intentions and executing data pre-processing.

Efficient Measurement of Service Similarity

The authors define a formal model for information services that incorporates the concept of service similarity. The model places services in metric spaces, and allows for services that have arbitrarily complex inputs and output domains. The authors then address the challenge of service substitution: finding the services most similar to a given service among a group, possibly large, of candidate services. To solve this nearest neighbor problem efficiently the authors embed the space of services into a vector space and search for the nearest neighbors in the target space. The authors report on an extensive experiment that validates both their formalization of similarity and their methods for finding service substitutions.