Transformers achieve great performance on Visual Question Answering (VQA). However, their systematic generalization capabilities, i.e., handling novel combinations of known concepts, is unclear. We reveal that Neural Module Networks (NMNs), i.e., question-specific compositions of modules that tackle a sub-task, achieve better or similar systematic generalization performance than the conventional Transformers, even though NMNs' modules are CNN-based. In order to address this shortcoming of Transformers with respect to NMNs, in this paper we investigate whether and how modularity can bring benefits to Transformers. Namely, we introduce Transformer Module Network (TMN), a novel NMN based on compositions of Transformer modules. TMNs achieve state-of-the-art systematic generalization performance in three VQA datasets, improving more than 30% over standard Transformers for novel compositions of sub-tasks. We show that not only the module composition but also the module specialization for each sub-task are the key of such performance gain.

This study addresses generating counterfactual explanations with multimodal information. Our goal is not only to classify a video into a specific category, but also to provide explanations on why it is not categorized to a specific class with combinations of visual-linguistic information. Requirements that the expected output should satisfy are referred to as counterfactuality in this paper: (1) Compatibility of visual-linguistic explanations, and (2) Positiveness/negativeness for the specific positive/negative class. Exploiting a spatio-temporal region (tube) and an attribute as visual and linguistic explanations respectively, the explanation model is trained to predict the counterfactuality for possible combinations of multimodal information in a post-hoc manner. The optimization problem, which appears during training/inference, can be efficiently solved by inserting a novel neural network layer, namely the maximum subpath layer. We demonstrated the effectiveness of this method by comparison with a baseline of the action recognition datasets extended for this task. Moreover, we provide information-theoretical insight into the proposed method.

This study aims at realization of electronic commerce (EC hereunder) through the Internet between Japan and China which have different systems. For this purpose, we have established the China Hosho Network (CHN hereunder) consortium which consists of universities and private research organizations. The consortium operates with an information disclosure policy, and is conducting a study on the credit guarantee system which will play a key role in facilitating the growth of EC between Japan and China [1].In EC, business is basically conducted on a credit basis; therefore many cases of failure are expected to occur in trade between Japan and China due to the difference in social and commercial systems and customs. Since the current civil codes of China do not provide regulations on international trade, there is a great demand among traders in Japan for establishment of credit guarantee systems for EC similar to damage insurance for international trade [2].This paper will discuss the development of the risk management system for EC between Japan and China, which functions as a credit guarantee system. The risk management system operates for its members who fulfil the terms of the membership agreement. It has a two-tire structure, consisting of risk control (preventive measures) and risk finance (financial arrangements including insurance coverage and/or other financial measures for possible trade failures). The risk control is a quality assurance system which can be realized by inspection before shipment. The risk finance is a guarantee insurance system which was developed incorporating conventional marine and fire insurance for trade. This paper also presents the results of a demonstrative application of the risk management system in order to prove its validity.