As machine learning technology is increasingly adopted into a variety of application domains, the potential risks of data leakage are becoming more serious in certain cases where the data contains highly sensitive information. While some privacy-preserving learning mechanisms for image data, such as SplitNN, enable the training of models without sharing private data on a central server, there exists a trade-off between security and computational cost to a client device. We propose a new mechanism to achieve higher level security and lower computational cost on a client device while maintaining model performance. Our approach, called Patch SplitNN, is based on SplitNN architecture that divides a CNN into two networks, called upper and lower. The difference from that previous work is to input individual image patches into multiple upper models, before concatenating their outputs before the lower model. For further improvement of the upper model training, we introduce an additional network and a loss function into the training process. We demonstrate our Patch SplitNN can classify images as accurately as a ResNet18 on various image classification datasets (CIFAR-10, CIFAR-100, and PCam) under multiple conditions (e.g. patching patterns, dropping patches).

In many applied computer vision tasks, training data is a scarce resource. This can result in poor performance of deep neural networks trained via supervised learning. We propose an approach to help boost performance in small training data regimes. Our method, called "CONGA", uses a Conditional GAN to Augment training data. Unlike previous work it can be added to any target discriminative model and allows the trade-off of computational cost for improved training sample efficiency. To further improve the quality of generated images and our method’s performance, and distinct from normal conditional GANs, we also propose to supervise the generator’s output via the target model. We compare our approach to similarly motivated methods on various image classification datasets (CIFAR-10, CIFAR-100, and Street View House Numbers), showing significant quantitative improvements.

A multi-fingered hand that can grasp and manipulate a variety of objects is an option for assisting people in their daily lives. However, the range of torque output that can be handled by the multi-fingered hand is very limited compared to the capability of the human hand. In this paper, we introduce a new multi-fingered hand consisting of a dynamic pulley and a linkage mechanism, aiming to achieve a human-like output torque with a human-like size. The proposed multi-fingered hand can achieve a fingertip force of 50N, which is equivalent to that of a human, and at the same time can perform delicate operations such as picking up a coin on a desk. In addition, we realized the stay-on-tab opening task of a can by utilizing fingertip strength.

A visual analysis method for finding a variety of human networks in chronicles on the basis of different viewpoints is proposed. A vast amount of human experience records is stored all over the world. For understanding and using such chronicles effectively, it is necessary to find several relations embedded in records. In this article, we propose a chronicle analysis tool (CAT), which provides several views of records from different viewpoints. For such purpose, we introduce the notion of chronicle data and a visual analysis framework for embedded relations in chronicles. On the basis of these notions, we describe a chronicle analysis tool and show a variety of historical human networks from different viewpoints as examples of its implementation.

This paper proposes methods of visualizing the personal promotion records of Japanese government officials in medieval times. Our research is aimed at developing multiple visualization tools that facilitate discovery of exceptional promotions as a way of highlighting interesting or important factors behind historical events, or that enable comparison of various promotion patterns to aid in understanding the political background the period. We have developed the following tools for visualizing personal promotion records:(i) the personal promotion graph (PPG-I) enables visualization of the promotional transition of court nobles in medieval Japan, and (ii) the promotion pattern graph (PPG-II) shows various promotion patterns that help reveal typical patterns of promotion for court nobles in relation to a particular period or family. There are many techniques for visualizing data in different ways. These tools provide variations in viewpoints for analyzing the political background of the era and help trigger consideration of different scenarios.

This paper proposes methods for extracting loanwords from Cyrillic Mongolian corpora and producing a Japanese-Mongolian bilingual dictionary. We extract loanwords from Mongolian corpora using our own handcrafted rules. To complement the rule-based extraction, we also extract words in Mongolian corpora that are phonetically similar to Japanese Katakana words as loanwords. In addition, we correspond the extracted loanwords to Japanese words and produce a bilingual dictionary. We propose a stemming method for Mongolian to extract loanwords correctly. We verify the effectiveness of our methods experimentally.

To generate content for multimedia encyclopedias, we propose a method for searching the Web, seeking images associated with a specific word sense. We use text in an HTML file that links to an image as a pseudo-caption for the image, enabling text-based indexing and retrieval. We use term descriptions in a Web search site called "Cyclone" as queries and match images and texts based on word senses. We show the effectiveness of our method experimentally.

We propose an associative document retrieval method, in which a document is used as a query to search for other similar documents. Because a long document usually includes more than one topic, we first analyze a query document to extract multiple subtopics. For each subtopic element, a sub-query is produced and similar documents are retrieved with a relevance score. The relevance scores are weighted by the importance of each subtopic element and are integrated to determine the final relevant documents. In the calculation of the subtopic importance, the specificity of a query term is evaluated using entropy, which is the deviation degree of the appearances of the term in each subtopic element. We apply this method to an invalidity patent search. By exploiting certain unique features of Japanese patent claims, we use features distinguishing the preamble and the essential portion in a query patent claim. To demonstrate the effectiveness of our method, we experimentally evaluated our associative document retrieval method on five years of patent documents.

We are developing a cross-media information retrieval system, in which users can view specific segments of lecture videos by submitting text queries. To produce a text index, the audio track is extracted from a lecture video and a transcription is generated by automatic speech recognition. In this paper, to improve the quality of our retrieval system, we extensively investigate the effects of adapting acoustic and language models on speech recognition. We perform an MLLR-based method to adapt an acoustic model. To obtain a corpus for language model adaptation, we use the textbook for a target lecture to search a Web collection for the pages associated with the lecture topic. We show the effectiveness of our method by means of experiments. 1.

We propose a cross-media lecture-on-demand system, in which users can selectively view specific segments of lecture videos by submitting text queries. Users can easily formulate queries by using the textbook associated with a target lecture, even if they cannot come up with effective keywords. Our system extracts the audio track from a target lecture video, generates a transcription by large vocabulary continuous speech recognition, and produces a text index. Experimental results showed that by adapting speech recognition to the topic of the lecture, the recognition accuracy increased and the retrieval accuracy was comparable with that obtained by human transcription. 1.

While recent retrieval techniques do not limit the number of index terms, out-ofvocabulary (OOV) words are crucial in speech recognition.Aiming at retrieving information with spoken queries, we fill the gap between speech recognition and text retrieval in terms of the vocabulary size.Given a spoken query, we generate a transcription and detect OOV words through speech recognition.We then correspond detected OOV words to terms indexed in a target collection to complete the transcription, and search the collection for documents relevant to the completed transcription.We show the effectiveness of our method by way of experiments.

Spoken queries submitted to question answering systems usually consist of query contents (e.g. about newspaper articles) and frozen patterns (e.g. WH-words), which can be modeled with N-gram models and grammar-based models, respectively. We propose a method to integrate those different types of models into a single N-gram model. We represent the two types of language models in a single word network. However, common smoothing methods, which are effective for N-gram models, decrease grammatical constraints for frozen patterns. For this problem, we propose a selective back-off smoothing method, which controls a degree to which smoothing is applied depending the network fragment. Additionally, resulting models are compatible with the conventional back-off N-gram models, and thus existing N-gram decoders can easily be used. We show the effectiveness of our method by way of experiments. 1.

We propose a method to generate large-scale encyclopedic knowledge, which is valuable for much NLP research, based on the Web. We first search the Web for pages containing a term in question. Then we use linguistic patterns and HTML structures to extract text fragments describing the term. Finally, we organize extracted term descriptions based on word senses and domains. In addition, we apply an automatically generated encyclopedia to a question answering system targeting the Japanese Information-Technology Engineers Examination.

In this paper, we propose a method to extract descriptions of technical terms from Web pages in order to utilize the World Wide Web as an encyclopedia. We use linguistic patterns and HTML text structures to extract text fragments containing term descriptions. We also use a language model to discard extraneous descriptions, and a clustering method to summarize resultant descriptions. We show the effectiveness of our method by way of experiments.

In this paper, we focus on the features of a lexicon for Japanese syntactic analysis in Japanese-to-English translation. Japanese word order is almost unrestricted and Kakuio-shi (postpositional case particle) is an important device which acts as the case label (case marker) in Japanese sentences. Therefore case grammar is the most effective grammar for Japanese syntactic analysis.The case frame governed by Yougen and having surface case (Kakuio-shi), deep case (case label) and semantic markers for nouns is analyzed here to illustrate how we apply case grammar to Japanese syntactic analysis in our system.The parts of speech are classified into 56 sub-categories.We analyze semantic features for nouns and pronouns classified into sub-categories and we present a system for semantic markers. Lexicon formats for syntactic and semantic features are composed of different features classified by part of speech.As this system uses LISP as the programming language, the lexicons are written as S-expression in LISP, punched onto tapes, and stored as files in the computer.