Young-Seob Jeong

An Intelligent Multi-Sensor Surveillance System for Elderly Care

This paper is an overview of our on-going project that proposes a monitoring system based on various sensors to detect risky situations for the elderly. From the standpoint of the end-user, a video surveillance system equipped with many other sensors can relieve caregivers from the need to keep a vigilant eye on each patient’s movements, while such technology can be effectively used for monitoring elderly people with dementia. Since a camera surveillance system has limits to classify complex human actions, this project aims to design an intelligent healthcare surveillance system, which extends the conventional automated video surveillance system with various additional sensors, to improve the performance of surveillance. The main contributions of our proposed system will be to: (i) minimize human intervention; (ii) detect more complex activities and situations using various sensors and improved sensor fusion techniques; and (iii) design a novel classifier that identifies risky situations with the collected information.

Sequential entity group topic model for getting topic flows of entity groups within one document

Topic mining is regarded as a powerful method to analyze documents, and topic models are used to annotate relationships or to get a topic flow. The research aim in this paper is to get topic flows of entities and entity groups within one document. We propose two topic models: Entity Group Topic Model (EGTM) and Sequential Entity Group Topic Model (S-EGTM). These models provide two contributions. First, topic distributions of entities and entity groups can be analyzed. Second, the topic flow of each entity or each entity group can be captured, through segments in one document. We develop collapsed gibbs sampling methods for performing approximate inference of the models. By experiments, we demonstrate the models by showing the analysis of topics, prediction performance, and the topic flows over segments in one document.

Weighted averaging fusion for multi-view skeletal data and its application in action recognition

Existing studies in skeleton-based action recognition mainly utilise skeletal data taken from a single camera. Since the quality of skeletal tracking of a single camera is noisy and unreliable, however, combining data from multiple cameras can improve the tracking quality and hence increase the recognition accuracy. In this study, the authors propose a method called weighted averaging fusion which merges skeletal data of two or more camera views. The method first evaluates the reliability of a set of corresponding joints based on their distances to the centroid, then computes the weighted average of selected joints, that is, each joint is weighted by the overall reliability of the camera reporting the joint. Such obtained, fused skeletal data are used as the input to the action recognition step. Experiments using various frame-level features and testing schemes show that more than 10% improvement can be achieved in the action recognition accuracy using these fused skeletal data as compared with the single-view case.

Temporal Information Extraction from Korean Texts

As documents tend to contain temporal information, extracting such information is attracting much research interests recently. In this paper, we propose a hybrid method that combines machine-learning models and hand-crafted rules for the task of extracting temporal information from unstructured Korean texts. We address Korean-specific research issues and propose a new probabilistic model to generate complementary features. The performance of our approach is demonstrated by experiments on the TempEval-2 dataset, and the Korean TimeBank dataset which we built for this study.

Discovery of research interests of authors over time using a topic model

With a growing number of Web documents, many approaches have been proposed for knowledge discovery on Web documents. The documents do not always provide keywords or categories, so unsupervised approaches are desirable, and topic modeling is such an approach for knowledge discovery without using labels. Further, Web documents usually have time information such as publish years, so knowledge patterns over time can be captured by incorporating the time information. In this paper, we propose a new topic model called the Author Topic-Flow (ATF) model whose objective is to capture temporal patterns of research interests of authors over time, where each topic is associated with a research domain. The design of the ATF model is based on the hypothesis that direct topic flows are better than indirect topic flows in the state-of-the-art Temporal Author Topic (TAT) model, which is the most similar approach to ours. We prove the hypothesis by showing the effectiveness of the ATF model compared to the TAT model.

Context awareness of social group by topic mining on visiting logs of mobile users in two dimensions

Ubiquitous computing technologies have been developed fast and various decision support systems were proposed. Consecutively, in recently, people are working on an implicit service agent of mobile phone to make people to be provided useful services without paying attentions. However, when people want to buy some products or to go somewhere, people are still going to use Internet to get preference information of group of people. They should put queries and click one or more documents. In this sense, how the implicit services could be perfectly available by only mining on the contextual data obtainable from smart phones? In this paper, we propose a model providing group preference information implicitly through topic-mining on the data obtainable from smart phones. We would introduce developed context awareness model in smart phones and apply a topic model which automatically learns a hierarchy of topics to our server. We would also show our simple experimentation to ensure the effectiveness of topic-mining and of provided implicit group preference information.

Discovery of topic flows of authors

With an increase in the number of Web documents, the number of proposed methods for knowledge discovery on Web documents have been increased as well. The documents do not always provide keywords or categories, so unsupervised approaches are desirable, and topic modeling is such an approach for knowledge discovery without using labels. Further, Web documents usually have time information such as publish years, so knowledge patterns over time can be captured by incorporating the time information. The temporal patterns of knowledge can be used to develop useful services such as a graph of research trends, finding similar authors (potential co-authors) to a particular author, or finding top researchers about a specific research domain. In this paper, we propose a new topic model, Author Topic-Flow (ATF) model, whose objective is to capture temporal patterns of research interests of authors over time, where each topic is associated with a research domain. The state-of-the-art model, namely Temporal Author Topic model, has the same objective as ours, where it computes the temporal patterns of authors by combining the patterns of topics. We believe that such ‘indirect’ temporal patterns will be poor than the ‘direct’ temporal patterns of our proposed model. The ATF model allows each author to have a separated variable which models the temporal patterns, so we denote it as ‘direct’ topic flow. The design of the ATF model is based on the hypothesis that ‘direct’ topic flows will be better than the ‘indirect’ topic flows. We prove the hypothesis is true by a structural comparison between the two models and show the effectiveness of the ATF model by empirical results.

Another dummy generation technique in location-based services

With the proliferation of mobile devices, many users now take advantage of location-based services that use their current position. However, careful consideration should be made when sending ones location to another as the location often includes personal attributes such as home address and reveals private information such as health or religion. To resolve this issue, a dummy generation technique is widely used. This technique protects the location privacy of a user by generating false position data (dummy) along with the true position data to obfuscate an adversary. However, the current dummy generation technique rarely assumes any prior knowledge held by the attacker that may allow them to reduce the level of uncertainty about the true location. In this paper, we propose a dummy generation method that is resistant to adversaries who have information about the user as well as external spatiotemporal knowledge. Our method uses conditional probabilities to generate realistic false locations at which the user is highly likely to be located at the given time and add more weight to the vulnerable location and time pairs. We first describe the strategy for the adversary and present our dummy generation method which is simple and effective for preventing the described attack. Experimental results show that our method obfuscates the true location more successfully compared to other approaches.

Semi-automated lifestyle manager for obesity

Existing healthcare services on smartphones are not convenient to use as they typically require too many manual inputs from users. This paper presents a semi-automated obesity-care application for helping people manage their own lifestyles. In our approach, a persons lifestyle is captured by a log of daily activities performed (e.g., eating, visiting, etc.), which can be captured automatically. For this purpose, the application recognizes the current location of the user and logs the history of his/her visiting places over time. This log can be used to deduce the type and the duration of the users activities and then the calorie consumption. The application also helps the user aware of his/her own lifestyle and BMI changes, and records this information for future usage such as when meeting a doctor for a clinical purpose. We have implemented the application on the Android platform and demonstrate the usefulness of this approach by illustrating the screen designs.

Money extraction and normalization from texts

Documents contain various types of information, and money information is one of such information. In the sentence “He borrowed ten dollars from me”, the expression ‘ten dollars’ conveys important information. When it is normalized into a specific way (e.g., 10 USD), then it can be used to develop various applications: Question-Answering (QA) system or Dialog system. In this paper, we propose an annotation language of the money information, and introduce a dataset of four languages: Korean, English, Chinese, and Japanese. We also discuss our module of extracting the money information, and show its performance by experiments.