Eisaku Maeda

Tokyo Denki University · Graduate School of Advanced Science and Technology

In the development of dialogue systems for android robots, the goal is to achieve human-like communication. However, subtle differences between android robots and humans are noticeable, leading even human-like android robots to be perceived differently. Understanding how humans accept android robots and optimizing their behavior is crucial. General...

In task-oriented dialogues with symbiotic robots, the robot usually takes the initiative in dialogue progression and topic selection. In such robot-driven dialogue, the user's sense of participation in the dialogue is reduced because the degree of freedom in timing and content of speech is limited, and as a result, the user's familiarity with and t...

Humans are easily conscious of small differences in an android robot's (AR's) behaviors and utterances, resulting in treating the AR as not-human, while ARs treat us as humans. Thus, there exists asymmetric communication between ARs and humans. In our system at Dialogue Robot Competition 2022, this asymmetry was a considerable research target in ou...

A dialogue system using an android robot has been open for our user study in the research area of communicative intelligence. In this paper, we introduce our counter sales robot system which were implemented various technique based on our experiences. The android which performs a travel agent was autonomously driven by our multimodal control such a...

There are many types of learning environments presented in higher education venues, requiring the development of a diverse repertoire of learning abilities. Group discussion (GD) is one such learning environment, and the students who participate require multiple communication skills. Technically, it is desirable for a student participating in a GD...

This paper describes an activity recognition method for the Sussex-Huawei Locomotion-Transportation (SHL) recognition challenge by Team TDU-DSML. The CNN model reported in our 2018 SHL Challenge was adopted. 5-second FFT spectrogram images from all axes of acceleration and gyro sensor data were treated as input data. We confirmed that a multiple-se...

For the Nurse Care Activity Recognition Challenge, an activity recognition algorithm was developed by Team TDU-DSML. A spatial-temporal graph convolutional network (ST-GCN) was applied to process 3D motion capture data included in the challenge dataset. Time-series data was divided into 20-second segments with a 10-second overlap. The recognition m...

In this paper, we propose a system based on Convolutional Long Short-Term Memory (ConvLSTM)-Attention Mechanism (AM) to preserve spatial features and time characteristics for surface electromyography (sEMG) signals. We assume that this method can perform more robustly under training with small data than a Convolutional Neural Network (CNN). To test...

Many research papers focus on communication skills in a group and information transmission capability of university students and/or business persons. Group discussions and poster session presentations are major topics in human interaction research. Aiming at benefits for students and teachers, we opened a living laboratory in our Tokyo Denki Univer...

At the SHL recognition challenge 2018, Team Tesaguri developed a human activity recognition method. First, we obtained the FFT spectrogram from 60-second acceleration and gyro sensor data for each of six axes. A five-second sliding window was used for FFT processing. About 70% of the spectrogram figures from the Sussex-Huawei Locomotion-Transportat...

To build a productive relationship between humans and artificial intelligence (AI), we must grasp the current situation as accurately as possible and make investments in the future toward developing such a relationship. This article introduces how we see and interpret the AI, Internet of Things, and big data era from the standpoint of promoting res...

The paradigm shift from information transmission to communication is taking place amid technological advancements in artificial intelligence (AI) and their attendant expectations. Against this background, I introduce NTT's AI-related research and development strategy and the research currently being pursued by NTT Communication Science Laboratories...

The era in which human beings are confronted with machines (computers or artificial intelligence) as disparate elements is coming to an end. From here on, we will embrace information science and technology as part of ourselves. This will necessitate the ability to decode, explore, and design the entire world, including us human beings. While bearin...

Basic research gives rise to innovation through new discoveries and inventions, and this can lead to changes in the structure of industry and our lifestyles. However, it is also true that such success stories are rare, and basic research carries a high degree of risk. This article analyzes the historical evolution of technologies born from the acti...

Research at NTT Communication Science Laboratories draws on both information science and human science with the aim of building a new technical infrastructure that will connect humans and information. These Feature Articles introduce new trends in the fields of speech, language, and hearing, which have a relatively long history of basic research.

This paper proposes hybrid dialogue control of both trigram and POMDP dialogue controls by extending our proposed method that uses two approaches: automatically acquiring POMDP structures and rewards for target dialogues through Dynamic Bayesian Networks (DBNs) with a large amount of dialogue data and reflecting action predictive probabilities into...

Partially Observable Markov Decision Processes (POMDPs) are applied in ac- tion control to manage and support users’ natural dialogue communication with conversational agents. Any agent’s action must be determined, based on probabilistic methods, from noisy data through sensors in the real world. Agents must flexibly choose their actions to reach a...

Canonical correlation analysis (CCA) is a powerful tool for analyzing multi-dimensional paired data. However, CCA tends to perform poorly when the number of paired samples is limited, which is often the case in practice. To cope with this problem, we propose a semi-supervised variant of CCA named "Semi CCA" that allows us to incorporate additional...

In this paper, which addresses smooth spoken interaction between human users and conversational agents, we present an experimental study that evaluates a method for user-adaptive coordination of agent communicative behavior. Our method adapts the pause duration preceding agent utterances and the agent gaze duration to reduce the discomfort perceive...

This paper presents an experimental study that analyzes how conversational agents activate human communication in thought-evoking multi-party dialogues be- tween multi-users and multi-agents. A thought-evoking dialogue, which is a kind of interaction in which agents act on user willingness to provoke user thinking, has the potential to stimulate mu...

From the viewpoint of supporting users' natural dialogue communication with conversational agents, their dialogue management has to determine any agent's action, based on probabilistic methods derived from noisy data through sensors in the real world. We believe unique Partially Observable Markov Decision Processes (POMDPs) should be applied to suc...

In order to design a dialogue system that users enjoy and want to be near for a long time, it is important to know the effect of the system's action on users. This paper describes ldquoWho is thisrdquo quiz dialogue system and its users' evaluation. Its quiz-style information presentation has been found effective for educational tasks. In our ongoi...

We propose a novel method for pose-invariant facial expression recognition from monocular video sequences that combines stochastic and determinis- tic search processes. We use the simple face model called variable-intensity template, which can be prepared with very little time and effort. We tackle the two issues found in previous work on the varia...

In this paper, we propose a method for pose-invariant facial expression recognition from monocular video sequences. The advantage of our method is that, unlike existing methods, our method uses a very simple model, called the variable-intensity template, for describing different facial expressions, making it possible to prepare a model for each per...

Our new research project called "ambient intelligence" concentrates on the creation of new lifestyles through research on communication science and intelligence integration. It is premised on the creation of such virtual communication partners as fairies and goblins that can be constantly at our side. We call these virtual communication partners mu...

This paper devises a novel kernel function for natural language processing tasks. The new kernels, called Hierarchical Directed Acyclic Graph (HDAG) kernels, directly accept graphs whose nodes could contain graphs. HDAG data structures are needed to fully reflect the syntactic and semantic structures inherently possessed by natural language data. I...

A research project on "ambient intelligence" recently launched by NTT Communication Science Laboratories aims to envision a new lifestyle made possible by communication science. Research and development of "ambient intelligence" should bridge the boundaries between technological fields and thus cover the entire field of communication science, rathe...

Docetaxel is one of the most effective anticancer drugs available in the treatment of breast cancer. Nearly half of the treated patients, however, do not respond to chemotherapy and suffer from side effects. The ability to reliably predict a patient's response based on tumor gene expression will improve therapeutic decision making and save patients...

In this paper, we address the problem of learning the order of a group of items. This study assumes that items have unobserved “scores” and are ranked by the score. The goal is to derive an approximate scoring function given ordered lists of items without their scores. We extend the notion of order statistics to any order and prove that the distrib...

We propose a method for assigning upper level gene ontology terms (GO categories) to genes using relevant documents. This method represents each gene as a vector using relevant documents to the gene. Then, binary classifiers are made for the GO categories using such supervised learning methods as support vector machines and maximum entropy method....

This paper proposes an interactive visualization method to support the exploration of data in decision-making and problem solving. Since this method employs an asymmetric communication mode, i.e. taking queries and requests expressed in a natural language as input and answering them with statistical charts, it can convert the normally tedious repet...

This paper devises a novel kernel function for structured natural language data. In the field of Natural Language Processing, feature extraction consists of the following two steps: (1) syntactically and semantically analyzing raw data, i.e., character strings, then representing the results as discrete structures, such as parse trees and dependency...

Convolution kernels, such as sequence and tree ker- nels, are advantageous for both the concept and ac- curacy of many natural language processing (NLP) tasks. Experiments have, however, shown that the over-fitting problem often arises when these ker- nels are used in NLP tasks. This paper discusses this issue of convolution kernels, and then propo...

This paper presents an application of Boosting for classifying labeled graphs, general structures for modeling a number of real-world data, such as chemical compounds, natural language texts, and bio sequences. The proposal consists of i) decision stumps that use subgraph as features, and ii) a Boosting algorithm in which subgraph-based decision st...

In this paper, we address the problem of statistical learnin g for multi- topic text categorization (MTC), whose goal is to choose all relevant top- ics (a label) from a given set of topics. The proposed algorithm, Max- imal Margin Labeling (MML), treats all possible labels as independent classes and learns a multi-class classifier on the induced m...

In this paper, we describe a method of automatic sentence alignment for building extracts from ab-stracts in automatic summarization research. Our method is based on two steps. First, we introduce the "dependency tree path" (DTP). Next, we calcu-late the similarity between DTPs based on the ESK (Extended String Subsequence Kernel), which con-siders...

It has become clear recently that there are many RNAs that are not translated into proteins, instead they work as functional molecules. These RNAs are called "noncoding RNAs." Predicting the secondary structure of these RNAs is important for understanding their functions. We focus on Nussinov's algorithm and the SCFG version of Nussinov's algorithm...

In this paper, we propose a machine learning-based method of multi-document summarization integrating sentence extraction with bunsetsu elimination. We employ Support Vector Machines for both of the modules used. To evaluate the effect of bunsetsu elimination, we participated in the multi-document summarization task at TSC-2 by the following two ap...

Introduction The largest amounts of information characteristic of the post-genomic era make the annotation of genes more and more important. Some projects have been launched to perform high-scale annotation using controlled vocabularies, such as the Gene Ontology Project[4]. However, these are constructed manually, which is very time consuming. Mor...

We have been investigating an interactive approach for Open-domain QA (ODQA) and have constructed a spoken interactive ODQA system, SPIQA. The system derives disambiguating queries (DQs) that draw out additional information. To test the efficiency of additional information requested by the DQs, the system reconstructs the user's initial question by...

This paper proposes a machine learning based question classification method using a kernel function, Hierarchical Directed Acyclic Graph (HDAG) Kernel. The HDAG Kernel directly accepts structured natural language data, such as several levels of chunks and their relations, and computes the value of the kernel function at a practical cost and time wh...

This paper proposes the "Hierarchical Directed Acyclic Graph (HDAG) Kernel" for structured natural language data. The HDAGKernel directly accepts several levels of both chunks and their relations, and then efficiently computes the weighed sum of the number of common attribute sequences of the HDAGs. We applied the proposed method to question classi...

Recently, open-domain question answering (ODQA) systems that extract an exact answer from large text corpora based on text input are intensively being investigated. However, the information in the first question input by a user is not usually enough to yield the desired answer. Interactions for collecting additional information to accomplish QA is...

this paper. 2 Question Answering Track Because of limited time, we applied our machine learning approach only to questions concerning dates and quantities; the other questions were processed in the same way as reported in [1]. Hereafter, we limit ourselves to the questions about dates and quantities

to summarization: (1) only sentence extraction, (2) sentence extraction + bunsetsu elimination. For both approaches, we use the machine learning algorithm called Support Vector Machines. We participated in both Task-A (single-document summarization task) and Task-B (multi-document summarization task) of TSC-2.

In this paper, we describe two Question Answering systems that participated in the NTCIR QAC task. The first system, SAIQA-Ii, follows a standard worddistance approach. This system was designed to output the top five candidates and participated in TASK-1. The second system, SAIQA-Is, employs a logic-based approach. This system was designed to outpu...

近年, インターネットや大容量の磁気デバイスの普及によって, 大量の電子化文書が氾濫している. こうした状況を背景として, 文書要約技術に対する期待が高まってきている. 特に, ある話題に関連する一連の文書集合をまとめて要約することが可能となれば, 人間の負担を大きく軽減することができる. そこで本稿では, 特定の話題に直接関連する文書集合を対象とし, 機械学習手法を用いることによって重要文を抽出する手法を提案する. 重要文抽出の手法としては近年, 自然言語処理研究の分野でも注目されている機械学習手法の1種であるSupport Vector Machineを用いた手法を提案する. 毎日新聞99年1年分より選んだ12話題の文書集合を用意し, それぞれの話題から総文数の10%, 30%, 50%...

500,000 PubMed abstracts. However, less than 50 documents are relevant for most queries. Applying scoring methods to all 500,000 abstracts would create a lot of noise. In the first step, we refined the document set with a simple keyword,search. For the second step, we developed two methods. The first method,(Method 1) uses a heuristic scoring syste...

This paper proposes an interactive approach to spoken interac-tive open-domain question answering (ODQA) systems. The goal of ODQA systems is to extract an exact answer to user's ques-tion from unstructured information sources such as large text cor-pora. When the reliabilities for answer hypotheses obtained by an ODQA system are low, systems need...

Question Answering (QA) is a technique that retrieves the correct answers from a large corpus to questions asked in natural language. QA, which has recently been extensively developed mainly in the United States, is an open-domain system and does not require a knowledge database. In this paper, we introduce a Japanese QA system, SAIQA, which has be...

A methodology is proposed for taking queries and requests expressed in natural language as input and answering them in charts through organizing that interaction into felicitous dialogue. Charts and graphics, as well as languages, are important modes of communication. This is especially true of those which are used frequently when people analyze hu...

A new pattern classification method called the Kernel-based Nonlinear Subspace (KNS) method is proposed. It implements a subspace method in a high-dimensional nonlinear space by a nonlinear transformation defined by kernel functions. The Support Vector Machine, a recent popular current research topic, is a nonlinear classification method employing...

This paper presents an answer selection method based on for Open-Domain Question Answering (QA). Selecting and ranking plausible answers from a large number of candidates in documents is one of the most critical parts of QA systems. It is extremely difficult to find good evaluation functions or rules for the answer selection. To overcome this issue...

Extracting sentences that contain important information from a document is a form of text summarization. The technique is the key to the automatic generation of summaries similar to those written by humans. To achieve such extraction, it is important to be able to integrate heterogeneous pieces of information. One approach, parameter tuning by mach...

In this report, we describe our question-answering system SAIQA-e (System for Advanced Interactive Question Answering in English) which ran the main task of TREC-10's QA-track. Our system has two characteristics (1) named entity recognition based on support vector ma- chines and (2) heuristic apposition detection. The MPR score of the main task is...

Humanoids, with human characteristics and functions, can act as mediators between human beings and information by means of information technology (IT) such as computers, networks, and sensors. We will be able to access various types of information instantaneously through the spread and development of IT. New generation humanoids will have a body wi...

The kernel based nonlinear subspace (KNS) method is proposed for multi-class pattern classification. This method consists of the nonlinear transformation of feature spaces defined by kernel functions and subspace method in transformed high-dimensional spaces. The support vector machine, a nonlinear classifier based on a kernel function technique, s...

Networks of cultured cortical neurones exhibit regular, synchronized, propagating bursts which are synaptically mediated, and which are hypothesized to play a part in activity-dependent formation of connections during development in vivo. The relationship between the strength of synaptic connections and the characteristics of synchronized propagati...

Experimental investigation of the dynamics of biological networks is a fundamental step towards understanding how the nervous system works. Spontaneous activity in cultured networks of cortical neurons has been investigated by using a multisite recording technique with planar electrode arrays. In these networks, the spatiotemporal firing patterns w...

Spontaneous activity in cultured networks of cortical neurons was investigated by extracellular recordings of their spatiotemporal firing patterns, using planar electrode arrays which comprised 64 microelectrodes. In the cultured cortical networks synchronous bursts states, asynchronous bursts states and non-burst firings states were observed by ch...

Long-term recording of spontaneous activity in cultured cortical neuronal networks was carried out using substrates containing multi-electrode arrays. Spontaneous uncorrelated firing appeared within the first 3 days and transformed progressively into synchronized bursting within a week. By 30 days from the establishment of the culture, the network...

The characteristics and mechanisms of synchronized firing in developing networks of cultured cortical neurons were studied using multisite recording through planar electrode arrays (PEAs). With maturation of the network (from 3 to 40 d after plating), the frequency and propagation velocity of bursts increased markedly (approximately from 0.01 to 0....

The characteristics and mechanisms of synchronized firing in developing networks of cultured cortical neurons were studied using multisite recording through planar electrode arrays (PEAs). With maturation of the network (from 3 to 40 d after plating), the frequency and propagation velocity of bursts increased markedly (approximately from 0.01 to 0....

Periodic synchronized bursts of action potentials in cultured neural networks of rat cortical neurons were recorded simultaneously at up to 16 sites using planar electrode arrays. The source of each burst was estimated from the relative delay of onset of activity between electrodes. The effects on network activity of electrical stimulation and of l...

An algorithm for object detection is presented. This method is robust against brightness variation and does not depend on the position, size, background pattern, shape, or color of target objects. The method consists of three processes: (1) normalize the brightness of the target and reference images; (2) calculate features, called the normalized pr...

This paper presents a method for creating a layered neural network that outputs stereo disparity maps, we call it the layered network for stereo(LNS). Disparity maps are obtained without iterative calculations since LNS accepts brightness values in stereo images and outputs stereo disparity maps of the input images. To train the LNS, we propose the...

We present a method which is robust to environmental conditions, that detects the existence of space filling objects such as cars, people, etc. This method is unaffected by uniform or local variations in image brightness induced by lighting conditions, weather, shadow of other objects such as buildings, trees, clouds, etc. Moreover, it does not dep...

A new general purpose method for object extraction and detection, RONPaC (Robust Object Extraction (Detection) using NPaC Features) method is presented. RONPaC employs normalized principal component (NPaC) features as a measure of similarity between corresponding regions of a target image and a background image. No a priori knowledge of objects and...

The general layered network for stereo (GLNS), a method for creating the general layered neural network to get stereo disparity maps, is presented. By using GLNS, disparity maps can be obtained without iterative calculations because it is a three-layered neural network. GLNS accepts gray levels in stereo images and output stereo disparity maps of t...