Gender recognition from unconstrained selfie images: a convolutional neural network approach

<p>Human gender recognition is an essential demographic tool. This is reflected in forensic science, surveillance systems and targeted marketing applications. This research was always driven using standard face images and hand-crafted features. Such way has achieved good results, however, the reliability of the facial images had a great effect on the robustness of extracted features, where any small change in the query facial image could change the results. Nevertheless, the performance of current techniques in unconstrained environments is still inefficient, especially when contrasted against recent breakthroughs in different computer vision research. This paper introduces a novel technique for human gender recognition from non-standard selfie images using deep learning approaches. Selfie photos are uncontrolled partial or full-frontal body images that are usually taken by people themselves in real-life environment. As far as we know this is the first paper of its kind to identify gender from selfie photos, using deep learning approach. The experimental results on the selfie dataset emphasizes the proposed technique effectiveness in recognizing gender from such images with 89% accuracy. The performance is further consolidated by testing on numerous benchmark datasets that are widely used in the field, namely: Adience, LFW, FERET, NIVE, Caltech WebFaces and<br />CAS-PEAL-R1.</p>

Cosine similarity-based algorithm for social networking recommendation

Social media have become a discussion platform for individuals and groups. Hence, users belonging to different groups can communicate together. Positive and negative messages as well as media are circulated between those users. Users can form special groups with people who they already know in real life or meet through social networking after being suggested by the system. In this article, we propose a framework for recommending communities to users based on their preferences; for example, a community for people who are interested in certain sports, art, hobbies, diseases, age, case, and so on. The framework is based on a feature extraction algorithm that utilizes user profiling and combines the cosine similarity measure with term frequency to recommend groups or communities. Once the data is received from the user, the system tracks their behavior, the relationships are identified, and then the system recommends one or more communities based on their preferences. Finally, experimental studies are conducted using a prototype developed to test the proposed framework, and results show the importance of our framework in recommending people to communities.

Machine Learning in Mobile Crowd Sourcing: A Behavior-Based Recruitment Model

With the advent of mobile crowd sourcing (MCS) systems and its applications, the selection of the right crowd is gaining utmost importance. The increasing variability in the context of MCS tasks makes the selection of not only the capable but also the willing workers crucial for a high task completion rate. Most of the existing MCS selection frameworks rely primarily on reputation-based feedback mechanisms to assess the level of commitment of potential workers. Such frameworks select workers having high reputation scores but without any contextual awareness of the workers, at the time of selection, or the task. This may lead to an unfair selection of workers who will not perform the task. Hence, reputation on its own only gives an approximation of workers’ behaviors since it assumes that workers always behave consistently regardless of the situational context. However, following the concept of cross-situational consistency, where people tend to show similar behavior in similar situations and behave differently in disparate ones, this work proposes a novel recruitment system in MCS based on behavioral profiling. The proposed approach uses machine learning to predict the probability of the workers performing a given task, based on their learned behavioral models. Subsequently, a group-based selection mechanism, based on the genetic algorithm, uses these behavioral models in complementation with a reputation-based model to recruit a group of workers that maximizes the quality of recruitment of the tasks. Simulations based on a real-life dataset show that considering human behavior in varying situations improves the quality of recruitment achieved by the tasks and their completion confidence when compared with a benchmark that relies solely on reputation.

OpenCarrier: Breaking the User Limit for Uplink MU-MIMO Transmissions With Coordinated APs

The global IoT market is experiencing a fast growth with a massive number of IoT/wearable devices deployed around us and even on our bodies. This trend incorporates more users to upload data frequently and timely to the APs. Previous work mainly focus on improving the up-link throughput. However, incorporating more users to transmit concurrently is actually more important than improving the throughout for each individual user, as the IoT devices may not require very high transmission rates but the number of devices is usually large. In the current state-of-the-arts (up-link MU-MIMO), the number of transmissions is either confined to no more than the number of antennas (node-degree-of-freedom, node-DoF) at an AP or clock synchronized with cables between APs to support more concurrent transmissions. However, synchronized APs still incur a very high collaboration overhead, prohibiting its real-life adoption. We thus propose novel schemes to remove the cable-synchronization constraint while still being able to support more concurrent users than the node-DoF limit, and at the same time minimize the collaboration overhead. In this paper, we design, implement, and experimentally evaluate OpenCarrier, the first distributed system to break the user limitation for up-link MU-MIMO networks with coordinated APs. Our experiments demonstrate that OpenCarrier is able to support up to five up-link high-throughput transmissions for MU-MIMO network with 2-antenna APs.

Mining High Utility Itemsets with Hill Climbing and Simulated Annealing

High utility itemset mining (HUIM) is the task of finding all items set, purchased together, that generate a high profit in a transaction database. In the past, several algorithms have been developed to mine high utility itemsets (HUIs). However, most of them cannot properly handle the exponential search space while finding HUIs when the size of the database and total number of items increases. Recently, evolutionary and heuristic algorithms were designed to mine HUIs, which provided considerable performance improvement. However, they can still have a long runtime and some may miss many HUIs. To address this problem, this article proposes two algorithms for HUIM based on Hill Climbing (HUIM-HC) and Simulated Annealing (HUIM-SA). Both algorithms transform the input database into a bitmap for efficient utility computation and for search space pruning. To improve population diversity, HUIs discovered by evolution are used as target values for the next population instead of keeping the current optimal values in the next population. Through experiments on real-life datasets, it was found that the proposed algorithms are faster than state-of-the-art heuristic and evolutionary HUIM algorithms, that HUIM-SA discovers similar HUIs, and that HUIM-SA evolves linearly with the number of iterations.

A New Approach for Mining Correlated Frequent Subgraphs

Nowadays graphical datasets are having a vast amount of applications. As a result, graph mining—mining graph datasets to extract frequent subgraphs—has proven to be crucial in numerous aspects. It is important to perform correlation analysis among the subparts (i.e., elements) of the frequent subgraphs generated using graph mining to observe interesting information. However, the majority of existing works focuses on complexities in dealing with graphical structures, and not much work aims to perform correlation analysis. For instance, a previous work realized in this regard, operated with a very naive raw approach to fulfill the objective, but dealt only on a small subset of the problem. Hence, in this article, a new measure is proposed to aid in the analysis for large subgraphs, mined from various types of graph transactions in the dataset. These subgraphs are immense in terms of their structural composition, and thus parallel the entire set of graphs in real-world. A complete framework for discovering the relations among parts of a frequent subgraph is proposed using our new method. Evaluation results show the usefulness and accuracy of the newly defined measure on real-life graphical datasets.

When Less Is More: Systematic Analysis of Cascade-Based Community Detection

Information diffusion, spreading of infectious diseases, and spreading of rumors are fundamental processes occurring in real-life networks. In many practical cases, one can observe when nodes become infected, but the underlying network, over which a contagion or information propagates, is hidden. Inferring properties of the underlying network is important since these properties can be used for constraining infections, forecasting, viral marketing, and so on. Moreover, for many applications, it is sufficient to recover only coarse high-level properties of this network rather than all its edges. This article conducts a systematic and extensive analysis of the following problem: Given only the infection times, find communities of highly interconnected nodes. This task significantly differs from the well-studied community detection problem since we do not observe a graph to be clustered. We carry out a thorough comparison between existing and new approaches on several large datasets and cover methodological challenges specific to this problem. One of the main conclusions is that the most stable performance and the most significant improvement on the current state-of-the-art are achieved by our proposed simple heuristic approaches agnostic to a particular graph structure and epidemic model. We also show that some well-known community detection algorithms can be enhanced by including edge weights based on the cascade data.

Social-Spatial Group Queries with Keywords

Social networks with location enabling technologies, also known as geo-social networks, allow users to share their location-specific activities and preferences through check-ins. A user in such a geo-social network can be attributed to an associated location (spatial), her preferences as keywords (textual), and the connectivity (social) with her friends. The fusion of social, spatial, and textual data of a large number of users in these networks provide an interesting insight for finding meaningful geo-social groups of users supporting many real-life applications, including activity planning and recommendation systems. In this article, we introduce a novel query, namely, Top- k Flexible Socio-Spatial Keyword-aware Group Query (SSKGQ), which finds the best k groups of varying sizes around different points of interest (POIs), where the groups are ranked based on the social and textual cohesiveness among members and spatial closeness with the corresponding POI and the number of members in the group. We develop an efficient approach to solve the SSKGQ problem based on our theoretical upper bounds on distance, social connectivity, and textual similarity. We prove that the SSKGQ problem is NP-Hard and provide an approximate solution based on our derived relaxed bounds, which run much faster than the exact approach by sacrificing the group quality slightly. Our extensive experiments on real data sets show the effectiveness of our approaches in different real-life settings.

Online Music Style Recognition via Mobile Computing

Music is a widely used data format in the explosion of Internet information. Automatically identifying the style of online music in the Internet is an important and hot topic in the field of music information retrieval and music production. Recently, automatic music style recognition has been used in many real life scenes. Due to the emerging of machine learning, it provides a good foundation for automatic music style recognition. This paper adopts machine learning technology to establish an automatic music style recognition system. First, the online music is process by waveform analysis to remove the noises. Second, the denoised music signals are represented as sample entropy features by using empirical model decomposition. Lastly, the extracted features are used to learn a relative margin support vector machine model to predict future music style. The experimental results demonstrate the effectiveness of the proposed framework.