Machine comprehension of text is the problem to answer a query based on a given context. Many existing systems use RNN-based units for contextual modeling linked with some attention mechanisms. In this paper, however, we propose StackReader, an end-to-end neural network model, to solve this problem, without recurrent neural network (RNN) units and its variants. This simple model is based solely on attention mechanism and gated convolutional neural network. Experiments on SQuAD have shown to have relatively high accuracy with a significant decrease in training time.

This abstract proposes a time series anomaly detector which 1) makes no assumption about the underlying mechanism of anomaly patterns, 2) refrains from the cumbersome work of threshold setting for good anomaly detection performance under specific scenarios, and 3) keeps evolving with the growth of anomaly detection experience. Essentially, the anomaly detector is powered by the Recurrent Neural Network (RNN) and adopts the Reinforcement Learning (RL) method to achieve the self-learning process. Our initial experiments demonstrate promising results of using the detector in network time series anomaly detection problems.

Social-based recommender systems have been recently proposed by incorporating social relations of users to alleviate sparsity issue of user-to-item rating data and to improve recommendation performance. Many of these social-based recommender systems linearly combine the multiplication of social features between users. However, these methods lack the ability to capture complex and intrinsic non-linear features from social relations. In this paper, we present a deep neural network based model to learn non-linear features of each user from social relations, and to integrate into probabilistic matrix factorization for rating prediction problem. Experiments demonstrate the advantages of the proposed method over state-of-the-art social-based recommender systems.

Most feature selection methods usually select the highest rank features which may be highly correlated with each other. In this paper, we propose a Stratified Feature Ranking (SFR) method for supervised feature selection. In the new method, a Subspace Feature Clustering (SFC) is proposed to identify feature clusters, and a stratified feature ranking method is proposed to rank the features such that the high rank features are lowly correlated. Experimental results show the superiority of SFR.

Fine-grained Entity Recognition (FgER) is the task of detecting and classifying entity mentions into more than 100 types. The type set can span various domains including biomedical (e.g., disease, gene), sport (e.g., sports event, sports player), religion and mythology (e.g., religion, god) and entertainment (e.g., movies, music). Most of the existing literature for Entity Recognition (ER) focuses on coarse-grained entity recognition (CgER), i.e., recognition of entities belonging to few types such as person, location and organization. In the past two decades, several manually annotated datasets spanning different genre of texts were created to facilitate the development and evaluation of CgER systems (Nadeau and Sekine 2007). The state-of-the-art CgER systems use supervised statistical learning models trained on manually annotated datasets (Ma and Hovy 2016). In contrast, FgER systems are yet to match the performance level of CgER systems. There are two major challenges associated with failure of FgER systems. First, manually annotating a large-scale multi-genre training data for FgER task is expensive, time-consuming and error-prone. Note that, a human-annotator will have to choose a subset of types from a large set of types and types for the same entity might differ in sentences based on the contextual information. Second, supervised statistical learning models when trained on automatically generated noisy training data fits to noise, impacting the model’s performance. The objective of my thesis is to create a FgER system by exploring an off the beaten path which can eliminate the need for manually annotating large-scale multi-genre training dataset. The path includes: (1) automatically generating a large-scale single-genre training dataset, (2) noise-aware learning models that learn better in noisy datasets, and (3) use of knowledge transfer approaches to adapt FgER system to different genres of text.

Thymio is a small, inexpensive, mass-produced mobile robot with widespread use in primary and secondary education. In order to make it more versatile and effectively use it in later educational stages, including university levels, we have expanded Thymio's capabilities by adding off-the-shelf hardware and open software components. The resulting robot, that we call Mighty Thymio, provides additional sensing functionalities, increased computing power, networking, and full ROS integration. We present the architecture of Mighty Thymio and show its application in advanced educational activities.

Data analysis education plays an important role in accelerating the efficient use of data analysis technologies in various domains. Not only the knowledge of statistics and machine learning, but also practical skills of deploying machine learning and data analysis techniques, are required for conducting data analysis projects in the real world. Data analysis competitions, such as Kaggle, have been considered as an efficient system for learning such skills by addressing real data analysis problems. However, current data analysis competitions are not designed for educational purposes and it is not well studied how data analysis competition platforms should be designed for enhancing educational effectiveness. To answer this research question, we built, and subsequently operated an educational data analysis competition platform called University of Big Data for several years. In this paper, we present our approaches for supporting and motivating learners and the results of our case studies. We found that providing a tutorial article is beneficial for encouraging active participation of learners, and a leaderboard system allowing an unlimited number of submissions can motivate the efforts of learners. We further discuss future directions of educational data analysis competitions.

Classifying products precisely and efficiently is a major challenge in modern e-commerce. The high traffic of new products uploaded daily and the dynamic nature of the categories raise the need for machine learning models that can reduce the cost and time of human editors. In this paper, we propose a decision level fusion approach for multi-modal product classification based on text and image neural network classifiers. We train input specific state-of-the-art deep neural networks for each input source, show the potential of forging them together into a multi-modal architecture and train a novel policy network that learns to choose between them. Finally, we demonstrate that our multi-modal network improves classification accuracy over both networks on a real-world large-scale product classification dataset that we collected from Walmart.com. While we focus on image-text fusion that characterizes e-commerce businesses, our algorithms can be easily applied to other modalities such as audio, video, physical sensors, etc.

Estimating economic and developmental parameters such as poverty levels of a region from satellite imagery is a challenging problem that has many applications. We propose a two step approach to predict poverty in a rural region from satellite imagery. First, we engineer a multi-task fully convolutional deep network for simultaneously predicting the material of roof, source of lighting and source of drinking water from satellite images. Second, we use the predicted developmental statistics to estimate poverty. Using full-size satellite imagery as input, and without pre-trained weights, our models are able to learn meaningful features including roads, water bodies and farm lands, and achieve a performance that is close to the optimum. In addition to speeding up the training process, the multi-task fully convolutional model is able to discern task specific and independent feature representations.

As the demand for mobile network services increases, immediate detection and forecasting of network failure events have become important problems for service providers. Several event detection approaches have been proposed to tackle these problems by utilizing social data. However, these approaches have not tried to solve event detection and forecasting problems from multiple data sets, such as web access logs and search queries. In this paper, we propose a machine learning approach that incorporates multiple user activity data into detecting and forecasting failure events. Our approach is based on a two-level procedure. First, we introduce a novel feature construction method that treats both the imbalanced label problem and the data sparsity problem of user activity data. Second, we propose a model ensemble method that combines outputs of supervised and unsupervised learning models for each data set and gives accurate predictions of network service outage. We demonstrate the effectiveness of the proposed models by extensive experiments with real-world failure events occurred at a network service provider in Japan and three user activity data sets.