Search

Data augmentation is a method to compensate a lack of sufficient amount of training data by increasing variations of the training data. It is also used even when there is a huge amount of training data to improve a generalization performance on the test data. In this paper, we propose a new method, Output-Augmentation (OA), which we use to improve the generalization performance without data augmentation. It augments each original output (but not input) and produces an arbitrary number of outputs which average to the original output. Updating the parameters is done by using the gradient over both the original and the augmented outputs. We conclude that the proposed novel method strongly complements the existing ones by showing empirical evaluations where we see improvements of the generalization performance in the task of image classification.

The mid-infrared range is an important spectrum range where materials exhibit a characteristic response corresponding to their molecular structure. A free-electron laser (FEL) is a promising candidate for a high-power light source with wavelength tunability to investigate the nonlinear response of materials. Although the self-amplification spontaneous emission (SASE) scheme is not usually adopted in the mid-infrared wavelength range, it may have advantages such as layout simplicity, the possibility of producing a single pulse, and scalability to a short-wavelength facility. To demonstrate the operation of a mid-infrared SASE FEL system in an energy recovery linac (ERL) layout, we constructed an SASE FEL setup in cERL, a test facility of the superconducting linac with the ERL configuration. Despite the adverse circumstance of space charge effects due to the given boundary condition of the facility, we successfully established the beam condition at the undulators and observed FEL emission at a wavelength of 20 μm. The results show that the layout of cERL has the potential for serving as a mid-infrared light source.

At the Photon Factory, we are constructing a new elliptically polarizing undulator(U#19) as a light source for soft X-ray scattering, spectroscopy and STXM experiments. U#19 is an EPU with an APPLE...

本研究では砂礫混合海浜で生じる分級や海浜断面の形成機構を明らかにするため, 水理模型実験を行って礫浜斜面上の流速場の計測を行うとともに, 礫浜の地形変化の計測を行って検討を加える. まず, 厚さの異なる一様勾配の礫浜の砕波帯の流速場の計測を行った. ついで移動床厚の異なる場合の海浜断面の波による変形と, 径や比重の異なる混合底質の移動形態と分級過程を観測した. また, 計測した流速場から底質の作用波力を評価し, 底質の移動限界時の力学的バランスから底質の移動特性を考究した. その結果, 礫浜斜面の断面形状には斜面内への浸透流が重要な役割を果たしていること, また, 底質の作用流体力と抵抗力の関係から底質の移動形態とその方向や形成される地形の概要が説明できることが判明した.