Deep Belief Network (DBN), which is well known to be a kind of Deep Learning methods, has a deep network architecture that can represent multiple features of input patterns hierarchically. Each layer employs a pre-trained Restricted Boltzmann Machine (RBM). In DBN model including RBMs, we may meet the difficulties in finding the optimal network structure and the best set of weights and threshold values. For the solution, we developed the adaptive structure learning method of DBN that can discover an optimal number of hidden neurons for given input data in a RBM by neuron generation / annihilation algorithm, and hidden layers in DBN by the extension of the algorithm. Moreover, the Long Short-Term Memory (LSTM) model can make an accurate prediction for a time series data set. The network architecture of LSTM has been proposed in various ways and can represent high accurate prediction for the benchmark data set, but the problems related to the optimal structure and some parameters still remains. In this paper, the adaptive structure model of RBM and DBN is applied to the LSTM model and the effectiveness was verified by 10-fold cross validation on benchmark data sets.

This paper proposes a multimodal genetic programming (GP) that incorporates a clustering of a population based on the tree structure similarity into GP and simultaneously acquires multiple local optimal solutions including a global optimal solution. The multimodal optimization problem aims to acquire not only a global optimal solution but also multiple local optimal solutions in a single optimization process. In general, although continuous real-valued optimizations are mainly targeted for multimodal optimization problems, problems with other solution structures, like a program in GP, have not been dealt with. This paper designs a multimodal program optimization problem that has a global and a local optimal solution and proposes a multimodal GP to acquires multiple local optimal programs including a global optimal one. Concretely, the proposed method separates the population into several clusters based on the similarity of tree structure, which is used as program expression in GP. Then, local optimum programs with different structure are acquired by optimizing each cluster separately. In order to investigate the effectiveness of the proposed method, we compare the proposed method with a simple GP without clustering on the designed multimodal GP benchmark. The experimental result reveals that the proposed method can acquire both the global and the local optimal programs at the same time.

This paper describes the vision model which consider with a gaze direction and optical flows while vehicle cornering. For developing a cornering assist system of automatically controlled vehicle, we focus on a relationship between gaze direction and optical flow based on a human visual perception. In this paper, we propose the visual model considering with a gaze direction and an optical flow as visual information using a spherical camera. An optical flow based on a spherical image is corresponding to a flow of light on the human retina. To verify the relationship between a spherical optical flow and gaze direction, we simulate the optical flow when a driver looks at the end of the curve. After that, we perform the actual driving experiment that human driving case and autonomous driving case using our prototype ultra-compact electric vehicle “TORiCLE”. According to the experimental results, we discuss about the explanation of the vehicle cornering by optical flow and the feasibility of a vehicle control system.

In recent years, deep convolutional neural network (DCNN) has widely been applied for image recognition, and shown a remarkable performance in various natural image-related applications. However, for medical image-related application such as computer-aided diagnosis (CAD), due to the limitation of training data and the modality difference between the natural and medical images, training the DCNN for medical image recognition is still a research topic. In this paper, we propose a DCNN-based method for lesion detection in mammograms. The proposed method consists of the following two steps. Given a mammogram, lesion candidates are firstly detected from the mammogram based on their intensity characteristics. Secondly, a transfer learning-based method is applied for training an existing DCNN to classify the lesion candidates into lesions or normal tissues. The proposed method is tested on a public mammogram database. Compared with several previous studies, our proposed method achieved a higher true positive rate and a lower false positive in lesion detection.

Recently, synthesis of gene regulatory networks having desired behavior has become of interest to many researchers and several studies have been done. We have proposed synthesis methods of gene regulatory networks in which desired behavior are given by expression pattern sequences. For realizing more complex behavior synthesizing more complex networks is needed. Realizing separatices in the state space of gene regulatory networks is helpful in order to synthesize gene regulatory networks having simpler structures. Therefore we have proposed a synthesis method for realizing separatrices. The purpose of this paper is to proof the existence of solutions to a synthesis problem of gene regulatory network for realizing separatrices. The proof is based on the procedure for finding a solution of the problem. Hence the proof also gives a synthesis method of gene regulatory networks for realizing separatirices.

We aim to estimate angle of finger during flexion and extension. Each finger joint angle is estimated by hybrid method. Moreover, it is verified the method is effectivity. The hybrid method is one of the method to identify finger motions by SVM and to estimate joint angle. In this research, an environment has been developed to be operate myoelectric prosthesis by myoelectric. Moreover, 5 finger motions have been identified by SVM, and it is estimated each finger joint angle of motion 1 with a high accuracy by the hybrid method. The myoelectric prosthesis has been operated based on these results, and it is affirmed the myoelectric prosthesis has been follow the operater. Therefore, the hybrid method is a usable way to estimate 5 finger joint angle. In the future, identification of finger motions will be increased accuracy, and finger angle of 2 motions will be estimated.

We have recently developed a portable light therapy apparatus for improvement of sleep quality and wakefulness. This technical note presents some preliminary experiments useful for design and evaluation of portable light therapy apparatus.