Continuously and unobtrusively identifying the phone's owner using accelerometer sensing and gait analysis has a great potential to improve user experience on the go. However, a number of challenges, including gait modeling and training data acquisition, must be addressed before unobtrusive gait verification is practical. In this paper, we describe a gait verification system for mobile phone without any assumption of body placement or device orientation. Our system uses a combination of supervised and unsupervised learning techniques to verify the user continuously and automatically learn unseen gait pattern from the user over time. We demonstrate that it is capable of recognizing the user in natural settings. We also investigated an unobtrusive training method that makes it feasible to acquire training data without explicit user annotation.

Abstract. Traditional graph-based clustering methods group vertices into non-intersecting clusters under the assumption that each vertex can belong to only a single cluster. On the other hand, recent research on graph-based clustering methods, applied to real world networks (e.g., protein-protein interaction networks and social networks), shows overlapping patterns among the underlying clusters. For example, in social networks, an individual is expected to belong to multiple clusters (or communities), rather than strictly confining himself/herself to just one. As such, overlapping clusters enable better models of real-life phenomena. Soft clustering (e.g., fuzzy c-means) has been used with success for network data as well as non-graph data, when the objects are allowed to belong to multiple clusters with a certain degree of membership. In this paper, we propose a fuzzy clustering based approach for community detection in a weighted graphical representation of social and biological networks, for which the ground truth associated to the nodes is available. We compare our results with a baseline method for both multi-labeled and single-labeled datasets.

Abstract. Network analysis has been an active area of research for the past few decades. Out of many open research questions that have been extensively studied, relational classification, community detection, link prediction are only to name a few. Collective classification is a well-known relational classification method for classifying entities (nodes) within a network which involves using both node based features and topological features of each node. It involves collective prediction of the unknown labels of all the test nodes in the network using label information of the training nodes. Even though this has been a well researched topic for years, very little has been done to address the following two challenges: (1) how to actively select the labeled nodes from the network to be used for training, and (2) how to efficiently obtain a sparse representation of the original network without losing much information, so that learning can scale to large networks. A lot of work has been done in theoretical computer science which aims towards finding the best approximation of large graphs. However, not much has been done from the perspective of finding an approximate subgraph that will help in classification of network datasets. In this paper, our contribution is in proposing an efficient graph sparsification method and a sampling technique which, along with the state-of-the-art network classifiers, can give comparable runtime and classification accuracies.

Abstract-Multi-label learning in graph-based relational data has gained popularity in recent years due to the increasingly complex structures of real world applications. Collective Classification deals with the simultaneous classification of neighboring instances in relational data, until a convergence criterion is reached. The rationale behind collective classification stems from the fact that an entity in a network (or relational data) is most likely influenced by the neighboring entities, and can be classified accordingly, based on the class assignment of the neighbors. Although extensive work has been done on collective classification of single labeled data, the domain of multi-labeled relational data has not been sufficiently explored. In this paper, we propose a neighborhood ranking method for multi-label classification, which can be further used in the Multi-label Collective Classification framework. We test our methods on real world datasets and also discuss the relevance of our approach for other multi-labeled relational data. Our experimental results show that the use of ranking in neighborhood selection for collective classification improves the performance of the classifier.

Machine learning and quantum computing are being progressively explored to shed light on possible computational approaches to deal with hitherto unsolvable problems. Classical methods for machine learning are ubiquitous in pattern recognition, with support vector machines (SVMs) being a prominent technique for network classification. However, there are limitations to the successful resolution of such classification instances when the input feature space becomes large, and the successive evaluation of so-called kernel functions becomes computationally exorbitant. The use of principal component analysis (PCA) substantially minimizes the dimensionality of feature space thereby enabling computational speed-ups of supervised learning: the creation of a classifier. Further, the application of quantum-based learning to the PCA reduced input feature space might offer an exponential speedup with fewer parameters. The present learning model is evaluated on a real clinical application: the diagnosis of Progressive Supranuclear Palsy (PSP) disorder. The results suggest that quantum machine learning has led to noticeable advancement and outperforms classical frameworks. The optimized variational quantum classifier classifies the PSP dataset with 86% accuracy as compared to conventional SVM. The other technique, a quantum kernel estimator, approximates the kernel function on the quantum machine and optimizes a classical SVM. In particular, we have demonstrated the successful application of the present model on both a quantum simulator and real chips of the IBM quantum platform.