Akito Sakurai

• Keio University

Current image quality assessment (IQA) methods require the original images for evaluation. However, recently, IQA methods that use machine learning have been proposed. These methods learn the relationship between the distorted image and the image quality automatically. In this paper, we propose an IQA method based on deep learning that does not req...

This research investigates vision‐based automated bridge component recognition, which is critical for automating visual inspection of bridges during initial response after earthquakes. Semantic segmentation algorithms with up to 45 convolutional layers are applied to recognize bridge components from images of complex scenes. One of the challenges i...

This paper proposes a quality recovery network (QRNet) that recovers the image quality from distorted images and improves the classification accuracy for image classification using these recovered images as the classifier inputs, which are optimized for image quality loss and classification loss. In certain image classification tasks, classifiers b...

Stagnant water on reinforced concrete (RC) decks shortens their fatigue life significantly compared to dry conditions. By using a multi-scale simulation together with the pseudo-cracking method, the remaining fatigue life of real RC bridge decks covered by stagnant water is estimated based upon their site-inspected surface crack's patterns. For qui...

In the midst of the increasing Port Structures that have crossed the in-service period of 50 years from construction, there is also deterioration in the number of divers carrying out the underwater investigation. Hence it is indispensible to develop a technique capable of carrying out wider area investigation in a shorter time, as an alternative of...

In this paper, we propose an image quality estimation method using deep learning. In the conventional image quality estimation method, image characteristics were analyzed for each distortion of the image, and a model was constructed. In recent years, image quality estimation method that learns automatically the relationship between distortion and i...

Randomized Artificial Crack Patterns are introduced to compensate the biased data of the real crack patterns from site.

By integrating a multi-scale simulation with the pseudo-cracking method, the remaining fatigue life of in-service reinforced concrete (RC) bridge decks can be estimated based upon their site-inspected crack patterns. But, it still takes time for computation. In order to achieve a quick deterioration-magnitude assessment of RC decks based upon their...

The goal of semi-supervised learning is to utilize many unlabeled samples under a situation where a few labeled samples exist. Recently, researches of semi-supervised learning are evolving with deep learning technology development, because, in deep, models have powerful representation to make use of abundant unlabeled samples. In this paper, we pro...

We generate a large number of predictive models by applying linear kernel SVR to historical currency rates’ bid data for three currency pairs obtained from high-frequency trading. The bid tick data are converted into equally spaced (1 min) data. Differences of price between the previous successive timeframes are used as features to predict the dire...

Many questions are posted on community websites in the world. Some of these questions are actually asked in order to receive empathy for the feelings of questioners, instead of getting specific answers to the questions asked. However, it is difficult to receive answers for these questions compared with questions that are asked for seeking responses...

We analyze the performance of various models constructed using linear kernel SVR and trained on historical bid data for high frequency currency trading. The bid tick data is converted into equally spaced (1 min) data. Different values for the number of training samples, number of features, and the length of the timeframes are used when conducting t...

Many questions are posted on community websites in the world. Some of these questions are actually asked in order to receive empathy for the feelings of questioners, instead of getting specific answers to the questions asked. However, it is difficult to receive answers for these questions compared with questions that are asked for seeking responses...

Widespread popularization of social networking services (SNSs) promoted, many studies about marketing activities using SNSs. But, there are a few studies focused on tendencies of consumer communities on SNSs. In this paper, we focus on a fashion brand and “Instagram”. And we study about tendencies of consumer communities on SNSs to present necessar...

Efficient market hypothesis is widely accepted in financial market studies and entails the unpredictability of future stock prices. In this study, we show that a simple analysis can classify short-term stock price changes with an 82.9% accuracy. Our analysis uses the order book information of high-frequency trading. The volume of high-frequency tra...

Widespread popularization of social networking services (SNSs) prompted, for example, a voluntary community such as an orchestra club of a university to use an SNS to support their activities. However, it is not all-purpose and lacks functions to improve members’ individual skills. Appropriate practice is a great help but we hardly find functions t...

In this paper, we propose a method to visualize information regarding hobbies and interests of a person inferred from tweets on Twitter to support informal communication in the real world. Analysis of the current states and experiments on informal communication clarified that it is important and useful for a person to know information such as hobbi...

This study targets social login registrants on an EC site and aims to clarify the difference between the purchasing tendency of social login registrants and general members by analyzing product purchasing history. We focused on the golf portal site that is the subject of this research. We analyzed the purchasing data comparing social login registra...

地名辞書の整備は地理情報システムを構築する上で重要な課題であるが，既存の辞書や手法では，口語的な地名表現 を自動的に地名と判定することは難しい.本研究では Twitter 文書内の単語の並びや構造に着目し，口語的な地名表現 を判定するシステムを提案する.大規模Twitterデータから構築したニューラルネットワーク言語モデル(NNLM)を 用いて，ベクトル空間内での既存地名への近接性から，ある単語が地名であるか否かを判定することを目指す.評価実 験を通じてNNLMとクラスタリングを組み合わせることで，口語的な地名を抽出できる可能性を提示した.

Recently, the number of voluntary communities such as local communities and university club activities are increasing. In these communities, since there are various types of members and there are no binding forces, it is usually difficult to maintain and im-prove member's motivation. To maintain and improve member's motivation, most of these commun...

In this study, we conducted foreign exchange rate simulated trading based on a traditional Japanese technical indicator called Ichimoku Kinkohyo, which is well-known and widely used in Japan for technical analysis on various kinds of market prices. We designed two trading strategies based on the support/resistance level of the five elements of Ichi...

Currency trading is an important area for individual investors, government policy decisions, and organization investments. In this study, we propose a hybrid approach referred to as MKL-DE, which combines multiple kernel learning (MKL) with differential evolution (DE) for trading a currency pair. MKL is used to learn a model that predicts changes i...

In this research, we built a support system for improving motivation by utilizing gamification, targeting one university circle, the Senshu University Philharmonic Orchestra, as an example of voluntary communities. The purpose of this research was to maintain and improve the motives of each individual orchestra member for practice. Analysis of the...

This study proposes a multiple kernel learning (MKL)-based regression model for crude oil spot price forecasting and trading. We used a well-known trend-following technical analysis indicator, the moving average convergence and divergence (MACD) indicator, for extracting features from original spot prices. Additionally, we factored in the possibili...

Predicting stock price change rates for providing valuable information to investors is a challenging task. Individual participants may express their opinions in social network service (SNS) before or after their transactions in the market; we hypothesize that stock price change rate is better predicted by a function of social network service activi...

In this paper, we propose a Laplacian minimax probability machine, which is a semi-supervised version of minimax probability machine based on the manifold regularization framework. We also show that the proposed method can be kernelized on the basis of a theorem similar to the representer theorem for non-linear cases. Experiments confirm that the p...

Recently, social games have attracted considerable attention. Building a relationship with other players can heighten the enjoyment derived from these games. However, many users play social games only with their fixed friends. There are functions to assist users to find friends in social games and SNS, but the existing functions are not simple enou...

This study applies a genetic algorithm (GA) to generate trading rules for currency trading based on a single technical indicator named the Relative Strength Index (RSI) as well as multiple timeframes from which we extract the feature. The target trading currency pair is EUR/USD and trading time horizon is one hour. Using more than one timeframe may...

Our proposed prediction and learning method is a hybrid referred to as MKL-GA, which combines multiple kernel learning (MKL) for regression (MKR) and a genetic algorithm (GA) to construct the trading rules. In this study, we demonstrate that the evaluation criteria used to examine the effectiveness of a financial market price forecasting method sho...

In trading in currency markets, reducing te mean of absolute or squared errors of predicted values is not valuable unless it results in profits. A trading rule is a set of conditions that describe when to buy or sell a currency or to close a position, which can be used for automated trading. To optimize the rule to obtain a profit in the future, a...

To recognize technical term, term dictionaries or tagged corpora are required, but it will take much cost to compile them. Moreover, the terms may have several representations and new terms may be developed, which complicates the problem further, that is, a simple dictionary building can’t solve the problem. In this research, to reduce the cost of...

This paper proposes a stock price prediction model, which extracts features from time series data and social networks for prediction of stock prices and evaluates its performance. In this research, we use the features such as numerical dynamics (frequency) of news and comments, overall sentiment analysis of news and comments, as well as technical a...

This paper proposes a stock price prediction model, which extracts features from time series data, news, and comments on the news, for prediction of stock price and evaluates its performance. In this research, we do not take account of text contents of news and user comments, but just consider numerical features of news and communication dynamics a...

This paper proposes a method to give an early warning of an abrupt change of price in a foreign exchange market. Volatility is a quantification of how much a value moves in a time series. It is now customary to assume that volatility of foreign exchange markets is time-varying. Intuitively we observe that there are at least two states or regimes: o...

In this paper we propose Manifold-Regularized Minimax Probability Machine, called MRMPM. We show that Minimax Probability Machine can properly be extended to semi-supervised version in the manifold regularization framework and that its kernelized version is obtained for non-linear case. Our experiments show that the proposed methods achieve results...

This paper proposes a hybrid model named MKL-GA, which combines Multiple Kernel Learning (MKL) and Genetic Algorithm (GA), for modeling and the prediction of FX (foreign exchange) rate on USDJPY currency pair by extracting features from three main FX pairs with three different short time horizons. Firstly, the MKL regression model predicts the chan...

The sensitivity of the M-cones and L-cones of an anomalous trichromat is lower than that of normal trichromats. By intensifying the light to the cones with lower sensitivity, the output ratio from the cone approached the level of a normal trichromat, and it was presumed that achieving close to original color recognition was a possibility. In order...

Naive Bayes (NB) is a simple Bayesian classifier that assumes the conditional independence and augmented NB (ANB) models are extensions of NB by relaxing the independence assumption. The averaged one-dependence estimators (AODE) is a classifier that averages ODEs, which are ANB models. However, the expressiveness of AODE is still limited by the res...

This paper states the issues faced, and the role played by keyword dictionaries with regard to discussion based web sites which aim to achieve a ’collective knowledge’ through the voluntary participation of corporate employees, and proposes a corrective strategy. A keyword dictionary is valuable in that it helps to integrate fragmented accumulated...

The sensitivity of the M-cones and L-cones of an anomalous trichromat is lower than that of normal trichromats. By intensifying the light to the cones with lower sensitivity, the output ratio from the cone approached the level of a normal trichromat, and it was presumed that achieving close to original color recognition was a possibility. In order...

It is meaningful to investigate the know-how of experienced project managers on the side of vendors about the risk in offshore software outsourcing. A survey is conducted to find out the main risk factors from the vendor’s viewpoint. The questions asked include background information of vendor and respondent, suggestions to the client, and evaluati...

Some of significant aspects associated with IT offshoring practices of several leading Japanese companies are discussed. Several companies, such as Toshiba, IBM-Japan, Hitachi, and Mitsubishi adopt IT offshore practices with their offshore vendors from countries, including China, Vietnam, India, Hungary, Romania, and Russia. Investigations reveal t...

Productivity is the key property of a natural language. Learnability is an equally important property, since productivity without learnability will not help the transfer of a language beyond generations. Linguistic productivity is supported by recursiveness described in terms of phrasal categories and systamaticity described in terms of lexical cat...

We propose a method of extracting named entities that are related to a single input word. Focusing on the syntactic dependency relation in sentences, it is reasonable to extract a case element that syntactically depends on the predicate that the input word depends on. In Japanese, though, a word which has appeared in a previous sentence is often om...

We propose an automatic and adaptive method to detect relatively important parts in multimedia content. The important parts of content should characterize the whole content and also be regarded as its summary. By applying our method, users capture important screenshots of the content and extract a set of important parts of the content that contain...

We give a necessary condition that a simple recurrent neural network with two sigmoidal hidden units to implement a recognizer of the formal language {a n b n | n > 0 } which is generated by a set of generating rules {S→aSb, S→ab } and show that by setting parameters so as to conform to the condition we get a recognizer of the language. The cond...

We propose in this paper a SOM-like algorithm that accepts online, as inputs, starts and ends of viewing of a multimedia content by many users; a one-dimensional map is then self-organized, providing an approximation of density distribution showing how many users see a part of a multimedia content. In this way “viewing behavior of crowds” informati...

As the volumes of software development increase and the cost reduction is required, most Japanese IT companies are interested in offshore software outsourcing. Although a lot of engineers have experienced the success and failure on their projects, their know-how still remains as tacit knowledge. This paper proposes a risk assessment scheme for new...

Data mining is the efficient discovery of patterns in large databases, and classification rules are perhaps the most important type of patterns in data mining applications. However, the number of such classification rules is generally very big that selection of interesting ones among all discovered rules becomes an important task. In this paper, fa...

Recent research hypothesizes that the capacity for syntactic recursions forms the computational core of a uniquely human language faculty. Contrary to this hypothesis, Gentner et al. claimed that the capacity to classify sequences from recursive, center-embedded grammar is not uniquely human. We show in this paper that the patterns Gentner used are...

We propose a new multimedia viewing environment in which users can effectively share their histories of viewing digital content. A viewing history is a record of which parts of the content the user viewed and/or listened to (and how many times) and which parts the user skipped. Since the parts viewed more frequently should reflect their importance,...

In this paper, we propose a method of extracting "action re- lations" between related topic words from Japanese weblogs (blogs). An action relation is a tuple of agent, target and predicate. Our method obtains blog articles that contain two keywords by AND search and outputs action relations con- structed from the predicate and the two keywords wit...

We experimentally show that, in the grammar learning by recurrent neural networks, the networks can learn a word categorization that is consistent with the grammatical categories of words in exemplar sentences.

We consider the problem of determining VC-dimension Ö{16( N \mathord/ \vphantom N logN logN )( 1 + o(1) ) + 4n2 } - 2n\sqrt {16\left( {{N \mathord{\left/{\vphantom {N {\log N}}} \right.\kern-\nulldelimiterspace} {\log N}}} \right)\left( {1 + o(1)} \right) + 4n^2 } - 2n and lower bounded by Ö{6( N \mathord/ \vphantom N logN logN )( 1 + o(1) )( 9...

We propose a method to improve the performance of R-learning, a reinforcement learning algorithm, by using multiple state-action value tables. Unlike Q- or Sarsa learning, R-learning learns a policy to maximize undiscounted rewards. Multiple state-action value tables cause substantial explorations as needed and make R-learning work well. Efficiency...

More sensors do not necessarily result in more appropriate state descriptions, so that a mobile robot has to select an appropriate set of sensors besides learning a state-action function in a reinforcement learning environment. We present a multi-armed bandit formulation of the problem and apply it to mobile robot navigation task. We modified the r...

We propose a role sharing model of language areas in which Broca’s area is for categorizing symbols used to represent rules stored and retrieved in other language areas. For example, at the syntactical level, the other language areas store rules represented with terminal symbols and also rules represented with non-terminal symbols, whereas Broca ar...

We propose a stochastic learning algorithm for multilayer perceptrons of linear-threshold function units, which theoretically converges with probability one and experimentally exhibits 100% convergence rate and remarkable speed on parity and classification problems with typical generalization accuracy. For learning the n bit parity function with n...

Our concept of a digital library from a user's viewpoint is explained by using a three-layer model of information distribution: information provider, information broker, and information user. Our worldwide communication-hall prototype consists of the virtual personal library client, called Webshelf, an information publishing server, and a hypermedi...

. We consider the VC-dimension of a set of the neural networks of depth s with w adjustable parameters that has h hidden units of activation functions of at most q segments of degree at most d polynomials. When d 2 and q 2, the VC-dimension is O(ws(s log d + log(qh=s))), O(ws((h=s) log q) + log d), andOmega ws log(dqh=s)). When d 2 and q = 1, it is...

. W 2 h 2 is an asymptotic upper bound for the VC-dimension of a large class of neural networks including sigmoidal, radial basis functions, and sigma-pi networks, where h is the number of hidden units and W is the number of adjustable parameters, which extends Karpinski and Macintyre's resent results.* The class is characterized by polynomial inpu...

0(ws(s log d+log(dqh/s))) and 0(ws((h/s) log q)+log[dqh/s)) are upper bounds for the VC-dimension of a set of neural networks of units with piecewise polynomial activation functions, where s is the depth of the network, h is the number of hidden units, w is the number of adjustable parameters, q is the maximum of the number of polynomial segments o...

We consider the problem of determining the VC-dimension δ3(h) of depth four n-input 1-output threshold circuits with h elements. Best known asymptotic lower bounds and upper bounds are proved, that is, when h → ∞, δ3(h) is upper bounded by ((h23) + nh)(log h)(1 + o(1)) and lower bounded by (12)((h24) + nh)(log h)(1 − o(1)). We also consider the pro...

The author shows that an n - h -1 artificial neural network with n real inputs, a single layer of h hidden units, and one binary output unit can store correctly at least n × h +1 examples in a general position. The proof is constructive so that weights are obtained deterministically from examples. The result is thought to be a generalization of the...

We propose an algorithm to prove a definite clause is modeled by an intended minimal model of a set of definite clauses. In the algorithm, no induction scheme will be used explicitly, although in effect a mathematical induction is implied. This algorithm corresponds to an inductionless induction method using the Knuth-Bendix algorithm in an equatio...

It has become popular to search the Internet or local databases for images. Meta information is commonly used in such image searches. It is, though, difficult and costly to annotate images with meta information that may describe their contents, which is also true in searching for paintings. As an important step for the content-based painting search...

We consider the problem of language identification in the limit from positive data enumerated by primitive recursive functions. Conditions are described under which families of languages are identifiable. These conditions depend on the behavior of inference machines when they are presented with enumerations of elements of unknown languages. How the...