Additional training of a deep learning model can cause negative effects on the results, turning an initially positive sample into a negative one (degradation). Such degradation is possible in real-world use cases due to the diversity of sample characteristics. That is, a set of samples is a mixture of critical ones which should not be missed and less important ones. Therefore, we cannot understand the performance by accuracy alone. While existing research aims to prevent a model degradation, insights into the related methods are needed to grasp their benefits and limitations. In this talk, we will present implications derived from a comparison of methods for reducing degradation. Especially, we formulated use cases for industrial settings in terms of arrangements of a data set. The results imply that a practitioner should care about better method continuously considering dataset availability and life cycle of an AI system because of a trade-off between accuracy and preventing degradation.

Systematic techniques to improve quality of deep neural networks (DNNs) are critical given the increasing demand for practical applications including safety-critical ones. The key challenge comes from the little controllability in updating DNNs. Retraining to fix some behavior often has a destructive impact on other behavior, causing regressions, i.e., the updated DNN fails with inputs correctly handled by the original one. This problem is crucial when engineers are required to investigate failures in intensive assurance activities for safety or trust. Search-based repair techniques for DNNs have potentials to tackle this challenge by enabling localized updates only on “responsible parameters” inside the DNN. However, the potentials have not been explored to realize sufficient controllability to suppress regressions in DNN repair tasks. In this paper, we propose a novel DNN repair method that makes use of the training history for judging which DNN parameters should be changed or not to suppress regressions. We implemented the method into a tool called Neurecover and evaluated it with three datasets. Our method outperformed the existing method by achieving often less than a quarter, even a tenth in some cases, number of regressions. Our method is especially effective when the repair requirements are tight to fix specific failure types. In such cases, our method showed stably low rates (<2 %) of regressions, which were in many cases a tenth of regressions caused by retraining.

Using computer technologies, the physical world in which we live and the virtual worlds generated by computers or popular cultures are coming closer together. Virtual agents are often designed that refer to humans in the physical world; at the same time, physical world products or services draw stories and emotions using characters. Computers can adjust virtual representations in the physical world based on the information they represent which is possessed by computers, and this adjustment leads to perceived personification. In this paper, we discuss the perception of personification and the possibility of application for persuasion. We developed a prototype application and then conducted surveys and a task-based user study. The results suggest the possibility of persuasion from personified agents superimposed close to an object.

Persuasive applications that change user behaviors and attitudes have been used for dietary support and healthcare and so on. When we use persuasive application to solve the social problems such as global warming and health issues, it is necessary that we extend the range of target users to people who are not interested in a problematic behavior. In this paper, we suggested an unconscious persuasive technique to design a persuasive application that is able to encourage people unconsciously. We have developed iDetective as a prototype application that persuades users unconsciously using agent persuasion and consciousness sensing. At the same time, we have evaluated the effectiveness of unconscious persuasion and influence on users through a user study using iDetective. As the result, although the unconscious persuasion on iDetective has yielded desirable effect, we could not find whether our application was able to attract people who are no interested in the target behavior. In addition, we have found that participants were not offended by the fact of unconscious persuasion and obtained possibilities and problems of unconscious persuasion.