著者
矢野 和洞 鈴木 丈裕 鈴木 智也
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.24, no.3, pp.113-122, 2020-05-15 (Released:2020-05-15)
参考文献数
10

Foreign-exchange (FX) brokers have some risk factors such as price fluctuation risk and latency of data transmission. To reduce these risks in FX brokerage services, we propose a short-term prediction of exchange rates quoted by counter-party banks. We consider that these exchange rates are generated by the knowledge of each counter-party bank, and therefore try to extract the knowledge by using a machine learning method. As a result, we could predict the direction of exchange rates with a prediction accuracy of about 80% if the prediction interval is 100[ms]. Furthermore, by integrating the knowledge of counterparty banks by the ensemble learning, we could improve not only prediction accuracy but also profitability of foreign-exchange brokers. These improvements can be considered as an effect of collective knowledge based on the diversity prediction theorem, but this effect might be limited by extremely short-term prediction of foreign-exchange rates after 100[ms]~200[ms].
著者
Naoya Murashima Hirokazu Kameoka Li Li Shogo Seki Shoji Makino
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.25, no.4, pp.145-149, 2021-07-01 (Released:2021-07-01)
参考文献数
24

This paper deals with single-channel speaker-dependent speech separation. While discriminative approaches using deep neural networks (DNNs) have recently proved powerful, generative approaches, including methods based on non-negative matrix factorization (NMF), are still attractive because of their flexibility in handling the mismatch between training and test conditions. Although NMF-based methods work reasonably well for particular sound sources, one limitation is that they can fail to work for sources with spectrograms that do not comply with the NMF model. To address this problem, attempts have recently been made to replace the NMF model with DNNs. With a similar motivation to these attempts, we propose in this paper a variational autoencoder (VAE)-based monaural source separation (VASS) method using a conditional VAE (CVAE) for source spectrogram modeling. We further propose an extension of the VASS method, called the discriminative VASS (DVASS) method, which uses a discriminative criterion for model training so that the separated signals directly become optimal. Experimental results revealed that the VASS method performed better than an NMF-based method, and the DVASS method performed better than the VASS method.
著者
He He Jun-Han Wang Shun Kojima Kazuki Maruta Chang-Jun Ahn
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.27, no.3, pp.49-57, 2023-05-01 (Released:2023-05-01)
参考文献数
16
被引用文献数
1

In a high-speed moving mobile environment, the channel state information (CSI) in the last part of the packet is different from the actual channel in the beginning part. Therefore, the channel estimation accuracy is degraded, especially when a small number of pilot symbols are used to ensure transmission efficiency. For the above reasons, it is necessary to compensate for CSIs to achieve reliable communication. Decision feedback channel estimation (DFCE) has been widely considered to be one of the channel tracking methods. However, the presence of time and frequency selective fading environments still causes estimation errors due to the decision-making process. We focused on the time-frequency domain response of the CSIs, which can be represented as a two-dimensional image. This paper newly proposes a regression convolutional neural network (CNN) based channel tracking scheme using the time-frequency domain response of the CSIs by DFCE for training and prediction to solve these problems. Computer simulation results demonstrate that the proposed scheme can achieve higher BER performance than the conventional schemes.
著者
Chisato Takahashi Kenya Jin'no
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.27, no.4, pp.65-68, 2023-07-01 (Released:2023-07-01)
参考文献数
5

Neural Architecture Search (NAS), which aims to automatically optimize the structure of a neural network for achieving excellent classification performance, has attracted considerable attention in recent years. Recently, zero-shot evaluation methods have been proposed for estimating classification performance without training to reduce the search time. However, these indices are still insufficient for finding the best-performing neural networks. In this study, we demonstrate that it is possible to evaluate convolutional neural networks (CNNs) using the robustness of the rectified linear unit (ReLU) output distribution to weights. We propose a new zero-shot CNN evaluation index based on this robustness index.
著者
Riki Watabe Hiroyuki Kamata
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.25, no.6, pp.227-231, 2021-11-01 (Released:2021-11-01)
参考文献数
8

In this paper, we propose a novel method for estimating the time delay in chaotic time series analysis. In recent years, focusing on the shape of an attractor using persistent homology has attracted attention. However, this method has a problem in that the calculation cost is enormous. In the proposed method, we aim to improve the calculation speed while considering the geometric shape of the attractor by focusing on the distance between the points in the data group.
著者
Reda Elbarougy Bagus Tris Atmaja Masato Akagi
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.24, no.6, pp.229-235, 2020-11-01 (Released:2020-11-01)
参考文献数
23

Speech and visual information are the most dominant modalities for a human to perceive emotion. A method of recognizing human emotion from these modalities is proposed by utilizing feature selection and long short-term memory (LSTM) neural networks. A feature selection method based on support vector regression is used to select the relevant features among thousands of features extended from speech and video features via bag-of-X-words. The LSTM neural networks then are trained using a number of selected features and also separately optimized for every emotion dimension. Instead of utterance-level emotion recognition, time-frame-based processing is performed to enable continuous emotion recognition using a database labeled for each time frame. Experimental results reveal that a system with feature selection is more effective for predicting emotion dimensions for a single language than the baseline system without feature selection. The performance is measured in terms of the concordance correlation coefficient obtained by averaging the valence, arousal, and liking dimensions.
著者
Takahiro Komai Song-Ju Kim Takuji Kousaka Hiroaki Kurokawa
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.23, no.4, pp.177-180, 2019-07-20 (Released:2019-07-20)
参考文献数
5
被引用文献数
2

In our previous studies, we showed that the estimation of the rock-scissors-paper (RSP, janken) game strategy is effective for the prediction of a player's hand sign sequences. The purpose of this study is to propose a method to estimate the RSP game strategy in the basis of human personality in an RSP game. To estimate a player's strategy in the RSP game, it is effective to compare the player's hand sign sequence and the hand sign sequences given by various typical RSP strategies on the basis of similarity. In this study, we propose the method of using a homology search to calculate the similarity between sequences. The results show that our proposed method is effective for strategy estimation.
著者
Yuki Hoshino Kenya Jin'no
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.22, no.4, pp.153-156, 2018-07-25 (Released:2018-07-25)
参考文献数
15
被引用文献数
1

Recently, machine learning has been attracting attention. Machine learning is mainly realized by the learning of artificial neural networks. Various learning methods have been proposed; however, the learning methods are based on gradient methods. On the other hand, swarm intelligence (SI) algorithms have been attracting attention in the optimization field. Generally speaking, SI algorithms have a large computation cost. Therefore, there are few cases of SI algorithms being applied to machine learning. In this paper, we propose a novel learning algorithm for an artificial neural network which applies our proposed nonlinear map optimization (NMO) method. NMO consists of some simple particles which are driven by a simple nonlinear map. NMO can be classified as an SI algorithm. However, it has only a small computation cost. Therefore, NMO can be applied to a learning algorithm for an artificial neural network. In this paper, we introduce NMO, and a small learning simulation is carried out to confirm the performance of our learning method.
著者
辻 広生 福水 洋平 道関 隆国 山内 寛紀
出版者
Research Institute of Signal Processing, Japan
雑誌
Journal of Signal Processing (ISSN:13426230)
巻号頁・発行日
vol.22, no.3, pp.121-134, 2018-05-25 (Released:2018-05-25)
参考文献数
13

We propose a multistructure convolutional neural network (CNN) for hiragana recognition of remarkably degraded license plate images captured by security cameras for the purpose of criminal investigation. The proposed multistructure CNN can use the optimal resolution image that cannot be used by conventional CNN by processing multiresolution images so that the recognition performance is improved. In many cases, plural candidates are allowed in remarkably degraded license plate character recognition for criminal investigation because it is not realistic to achieve practical level correct rate with a single candidate. The general criterion of practical level recognition accuracy for criminal investigation is whether the method achieves the correct rate of 90 percent by allowing up to the second candidate. Generally, the recognition accuracy of CNN decreases when the degradation estimation is inaccurate, and the CNN is not optimized. Under the condition that the CNN was not optimized, the proposed multistructure CNN could achieve practical level recognition performance while the conventional CNN could not achieve that performance.