Transactions of the Japanese Society for Artificial Intelligence Vol. 32(2017) No. 4

This paper proposes a novel neural networks based model for learning user-independent features. In activity recognition using wearable sensors, user-independence of features could provide better user-generalization performance, enhance privacy protection, and both are important for using activity recognition techniques in a real-world scenario. However, designing such features is not an easy task, because it is not clear what kind of features become user-independent, and moreover, poor design of user-independence harms activity recognition performance.Hear, we propose User-Adversarial Neural Networks for automatically learning user-independent features. The proposed model considers an adversarial-user classifier in addition to a regular activity classifier in the training phase, and learn the features that help to distinguish the activities but obstruct to distinguish the users. In other words, the model explicitly penalizes representations for becoming user-dependent, while keeping activity recognition performance as much as possible. Our main result is an empirical validation on three activity recognition tasks regarding wearable sensor based activity recognition. The result shows the proposed model improves independence of features comparing with the regular deep convolutional neural networks in both qualitatively and quantitively. We also summarize future work for better user-generalization and privacy protection from the perspective of the representation learning.

Nearly a hundred years ago, Karl von Frisch discovered that honeybees (Apis mellifera) communicate the exact location of food sources to other bees through a complex movement called waggle dance. Since then, analyzing communications performed by honeybee workers in their hive is one of the most important and interesting issues to reveal a mechanism of honeybee's language. However, it is not clear yet that the behavioral developmental process of young adult honeybees after their emerging adulthood. Our research focus is to analyze how they learn to dance well. These analyses have been usually conducted by extracting honeybee ’s walking trajectories from recorded long-time video data manually. For a systematic and theoretical analysis of honeybee's communication, we have developed an automatic tracking algorithm of multiple honeybees using image processing and constructed an automatic recording system for long-term tracking of honeybee behaviors with Radio Frequency Identification (RFID) sensors and high-resolution camera modules using multiple small-size single board computers, RaspberryPi. The tiny RFID-tags are mounted on the dorsal tergum of young adult honeybees just after their emergence and two RFID antennas are arranged about 20 centimeters apart to determine the time whether each honeybee was entering or leaving the hive. Using this system, we conducted a recording experiment from 6:30 am to 7:30 pm over 4 weeks in September 2015. The size of our target colony is about 400 honeybees including a queen. The number of tagged honeybees is 100 individuals and the printed numbers are attached for each to be identified by observers. In this paper, first we show the background of our research and clarify requirements for a monitoring system of honeybee behaviors. Next, we show an overview of our system and explain the simultaneous tracking algorithm we proposed. Finally, we show the experimental results and discuss the system capabilities and some open problems.

The amount of costs for long-term care is increasing because of aging society. It is important to improve quality of care process. We focus on sharing the knowledge of care process in care facility to achieve the goal. There are some issues to share the knowledge. (1) The knowledge is implicit on the care workers. (2) It is difficult to unite the knowledge as world standard because of its variety. We proposed the methodology of "knowledge explication" to overcome these issues. The methodology is (1) to explicate the knowledge of care process by employees in care facility. (2) It is performed on each care facility so they can explicate the facility-specific knowledge of care process. We applied the methodology to two care facilities and evaluate it. The explicated knowledge is 1.8 times more than the knowledge from textbook. We also confirmed the efficiency of the methodology by questionnaire to the participants.

Learning distributed representations for relation instances is a central technique in downstream NLP applications. In particular, semantic modeling of relations and their textual realizations (relational patterns) is important because a relation (e.g., causality) can be mentioned in various expressions (e.g., “X cause Y”, “X lead to Y”, “Y is associated with X”). Notwithstanding, the previous studies paid little attention to explicitly evaluate semantic modeling of relational patterns. In order to address semantic modeling of relational patterns, this study constructs a new dataset that provides multiple similarity ratings for every pair of relational patterns on the existing dataset [Zeichner12]. Following the annotation guideline of [Mitchell 10], the new dataset shows a high inter-annotator agreement. We also present Gated Additive Composition (GAC), which is an enhancement of additive composition with the gating mechanism for composing distributed representations of relational patterns. In addition, we conduct a comparative study of different encoders including additive composition, RNN, LSTM, GRU, and GAC on the constructed dataset. Moreover, we adapt distributed representations of relational patterns for relation classification task in order to examine the usefulness of the dataset and distributed representations for a different application. Experiments show that the new dataset does not only enable detailed analyses of the different encoders, but also provides a gauge to predict successes of distributed representations of relational patterns in the relation classification task.

The present study investigated the influence of reflections on self/others’ trust within group-based problem solving. The study assessed the role of trust dynamics on perspective-taking activities within conflictive groups, extending the experimental framework used by a previous study and including conversational agents for controlling participants’ interactions related to trust dynamics and perspective taking behavior. Results showed that (1) reflections of self/other trust in conflictive groups may influence trust towards other members, and (2) reflections of trust by members with conflicting perspectives may facilitate trust and perspective taking process. This suggests that the level of trust dynamics facilitates trust and can function to manifest perspective taking within cooperative groups. The results of the study provide new knowledge in collaborative problem solving studies that the development of trust has a progressive effect on perspective taking activities among conflictive members.