Recently, evaluation of Quality of Life (QoL) in everyday life has been regarded as important. It will be possible to provide in-home services which improve QoL if the stress level of the dweller while performing daily activities (e.g., cleaning, cooking, etc.) which take temporal and physical burden is recognized during daily life. In this study, we analyze the stress level while performing daily activities by the LF/HF ratio known as a stress evaluation index based on heartbeat. As an experiment, we asked five participants to live in our smart home testbed wearing wearable device (Empatica E4 wristband) for four days each, and collected biometric data including heartbeat and volume pulse wave while recording activity labels. Through analysis of the collected data, we report the relation between the daily living activities and the stress index.

Declining birthrate and aging population are common problems of developed countries. These cause increasing of single life of the elderly, and finally, increase the risk of lonely death of them. One of the best precautions against lonely death is the safety confirmation by a third person. So, the service of that for the elderly is spreading in various forms now. However, the service is not popular because of some kinds of problems, for example, workload, technical difficulty, and privacy. Therefore, the method using ambient sensors in the home has been proposed. In this method, some kinds of sensors are installed in living environment, such as motion sensor and door sensor and so on. These sensors have no personal information. The data is collected by sensors periodically and observed for safety confirmation. However, the sensor data have different meanings by lifestyle of residents. So, to determine the standard of safety, we have to consider the difference of lifestyle according to some conditions such as person and season. Hence, we aim to develop a new safety confirmation system which can optimize safety standard according to the resident's lifestyle. In this paper, we collected the sensor data of 19 elderly people for about 10 months and analyzed it as a first step of developing. As a result, we found the difference of sensor data in lifestyles according to residents and seasons.

飲食店向け不動産物件の賃料は，不動産会社のベテラン営業職員が培ってきた経験や勘といった暗黙知に基づいて決定されている．賃料の決定要因としては，物件固有の情報である静的情報，物件周辺の情報である動的情報，物件の特徴などを含む潜在的情報が挙げられている．潜在的情報は，ベテラン営業職員による指標化が難しいとされる情報である．本研究では潜在的情報として，物件に付与されているキャッチコピーを用いる手法を提案する．キャッチコピーはDoc2Vec によりベクトル表現に変換し，ノイズを除去するため，品詞の選別を行なった．その結果，品詞の選別を行うことで推定精度が向上し，重回帰分析を用いた場合に，決定係数が0.611 と最も高い値が得られた．

企業や家計などの経済主体の活動は，様々なメカニズムにより，相互に影響を与えていく．そのため，先行的な活動をとらえた経済統計には，先行きの経済の動きを示唆する情報が含まれると考えられる．本研究では，内閣府景気動向指数の先行指数の算出に採用される11の先行系列を用いて，各種機械学習手法を用いた短期経済予測モデルを構築し，その予測特性の検証を行った．その結果，ディープラーニング手法の一種であり再帰的なネットワーク構造を持つRNN（リカレントネットワークニューラルネットワーク）において，相対的に高い予測精度が確認できた．また，RNNベースの予測モデルの構築の際，景気動向指数の先行指数のみを学習に用いた場合に比べて，先行系列11系列を学習に用いた場合，先行きの経済予測の精度に改善の傾向が見られた．このことは，経済的に先行性を持つと解釈される複数の系列の情報を，RNNを用いて直接モデル化することで，先行きの経済予測に関して有用な情報の抽出が可能となることを示唆する結果と言えよう．

The evaluation function for an imperfect information game is always hard to define but has a significant impact on the playing strength of a program. Deep learning has made great achievements in several recent years, and already exceeded the level of top human players in perfect information games such as AlphaGo. Predicting opponents moves and hidden states is important in imperfect information games. This paper describes a model on building a Mahjong artificial intelligence with deep learning method and supervised learning theory. Four deep neural network for discarding and predicting opponents' waiting, waiting tiles and point changes are combined into one model and performs good during games. With improved feature engineering, our accuracies on validation data of these networks reach higher than Dr. Mizukami and Professor Tsuruoka's network.

In a large scale IoT system, structured and unstructured data are collected from various distributed sensors. It is important to visualize these data with an overview context composed of multiple views and interactively focus on some detail views to understand the current system status. But a unified VA (Visual Analysis) is difficult owing to due to being short of expressly relationship between distributed datasets from different sensors or processed subsets of big data. In this paper, we present a visualization framework to analytically acquire the relationship among distributed or processed subsets, integrate their views in a visualization context, and realize operation linkage between them.