Development of training environment for deep learning with medical images on supercomputer system based on asynchronous parallel Bayesian optimization

The Journal of Supercomputing - Recently, deep learning has been exploited in the field of medical image analysis. However, the training of deep learning models with medical images is...     

Human-centered image classification via a neural network considering visual and biological features

Multimedia Tools and Applications - In this paper, we propose a human-centered image classification via a neural network considering visual and biological features. The proposed method has two...     

Head direction estimation from low resolution images with scene adaptation

This paper presents an appearance-based method for estimating head direction that automatically adapts to individual scenes. Appearance-based estimation methods usually require a ground-truth dataset taken from a scene that is similar to test video sequences. However, it is almost impossible to acquire many manually labeled head images for each scene. We introduce an approach that automatically aggregates labeled head images by inferring head direction labels from walking direction. Furthermore, in order to deal with large variations that occur in head appearance even within the same scene, we introduce an approach that segments a scene into multiple regions according to the similarity of head appearances. Experimental results demonstrate that our proposed method achieved higher accuracy in head direction estimation than conventional approaches that use a scene-independent generic dataset.     

Experiments on thermally driven acoustic oscillations of gaseous helium

This paper describes experimental studies of thermally driven acoustic oscillations of a gas column (Taconis vibration) generated in a pipe whose end is closed at the warm part and open at the cold part. The stability curves and the frequency diagrams of the oscillations are experimentally determined under a given temperature distribution for the ratio of warm length of pipe to cold length as a parameter. The existence of two branches as predicted by Rott's theory is confirmed. A finite boundary layer thickness plays an important role in exciting and characterizing this type of instability.     

A photodiode amplifier system for pulse-by-pulse intensity measurement of an x-ray free electron laser

We have developed a single-shot intensity-measurement system using a silicon positive-intrinsic-negative (PIN) photodiode for x-ray pulses from an x-ray free electron laser. A wide dynamic range (10(3)-10(11) photons/pulse) and long distance signal transmission (>100 m) were required for this measurement system. For this purpose, we developed charge-sensitive and shaping amplifiers, which can process charge pulses with a wide dynamic range and variable durations (ns-μs) and charge levels (pC-μC). Output signals from the amplifiers were transmitted to a data acquisition system through a long cable in the form of a differential signal. The x-ray pulse intensities were calculated from the peak values of the signals by a waveform fitting procedure. This system can measure 10(3)-10(9) photons/pulse of ~10 keV x-rays by direct irradiation of a silicon PIN photodiode, and from 10(7)-10(11) photons/pulse by detecting the x-rays scattered by a diamond film using the silicon PIN photodiode. This system gives a relative accuracy of ~10(-3) with a proper gain setting of the amplifiers for each measurement. Using this system, we succeeded in detecting weak light at the developmental phase of the light source, as well as intense light during lasing of the x-ray free electron laser.     

Robust l ∞ preview control for biped walking pattern generation

We propose a robust l _∞ preview control for the pattern generation of biped walking. First, we give the robust l _∞ preview control robustness against the initial value and apply it to the cart-table model. Next, we propose a method to reduce the conservativeness of the robust l _∞ preview control and an algorithm to solve it. Finally, the effectiveness of our proposed method is illustrated by a simulation.     

A novel genetic algorithm with different structure selection for circuit design optimization

In the traditional GA, the tournament selection for crossover and mutation is based on the fitness of individuals. This can make convergence easy, but some useful genes may be lost. In selection, as well as fitness, we consider the different structure of each individual compared with an elite one. Some individuals are selected with many different structures, and then crossover and mutation are performed from these to generate new individuals. In this way, the GA can increase diversification into search spaces so that it can find a better solution. One promising application of GA is evolvable hardware (EHW), which is a new research field to synthesize an optimal circuit. We propose an optimal circuit design by using a GA with a different structure selection (GAdss), and with a fitness function composed of circuit complexity, power, and signal delay. Its effectiveness is shown by simulations. From the results, we can see that the best elite fitness, the average fitness value of correct circuits, and the number of correct circuits with GAdss are better than with GA. The best case of optimal circuits generated by GAdss is 8.1% better in evaluation value than that by traditional GA.     

Numerical analysis of feedforward position control for non-pulley musculoskeletal system: a case study of muscular arrangements of a two-link planar system with six muscles

In a musculoskeletal system like a tendon-driven robot, redundant actuation is necessary because muscles (or mechanical parts such as tendons) can transmit tension only unidirectionally. This redundancy yields internal force among muscles, which has a particular field of potential energy. Using internal force as a feedforward input, a musculoskeletal system can achieve feedforward position control with no sensory feedback. This paper studies the feedforward position control coming from the redundancy for a non-pulley musculoskeletal system. Targeting a planar two-link system with six muscles as a case study, the motion convergence depending on the muscular arrangement is examined quasi-statically. The results point out that the convergence is extremely sensitive to the muscular arrangement, and adding small offsets for the muscular connected points can remarkably improve the positioning performance.     

A novel tumor-related protein, C7orf24, identified by proteome differential display of bladder urothelial carcinoma

Proteome analysis of bladder cancer with narrow-range pH 2-DE has identified a novel protein on chromosome 7 encoded by ORF 24 (C7orf24) as one of the highly expressed proteins in cancer cells. C7orf24 is currently registered in the protein database as a hypothetical protein with unknown function. The homologs of C7orf24 in other animals have also been registered as putative protein genes. Western blot analysis using a mAb against C7orf24 confirmed its higher expression in bladder cancer compared with normal tissue. Several other cancer cell lines were also found to express C7orf24. However, the introduction of C7orf24 into Rat-1 or NIH3T3 cells did not cause malignant transformation. A stable transfectant of NIH3T3 cells with recombinant retrovirus vector was produced for a growth rate assay, and a higher growth rate was observed in C7orf24-expressing cells compared with the controls. Six kinds of small interfering RNAs (siRNAs) were then produced, and C7orf24-siRNA#5 showed a strong knockdown effect on protein expression and significant antiproliferative effects on cancer cell lines were demonstrated by the MTT assay. Therefore, C7orf24 may have an important role in cancer cell proliferation, and may be an appropriate therapeutic target molecule against cancer.