Some neural network can be trained by transfer learning, which uses a pre-trained neural network as the source task, for a small target task’s dataset. The performance of the transfer learning depends on the knowledge (i.e., layers) selected from the pre-trained network. At present, this knowledge is usually chosen by humans. The transfer learning method PathNet automatically selects pre-trained modules or adjustable modules in a modular neural network. However, PathNet requires modular neural networks as the pre-trained networks, therefore non-modular pre-trained neural networks are currently unavailable. Consequently, PathNet limits the versatility of the network structure. To address this limitation, we propose Stepwise PathNet, which regards the layers of a non-modular pre-trained neural network as the module in PathNet and selects the layers automatically through training. In an experimental validation of transfer learning from InceptionV3 pre-trained on the ImageNet dataset to networks trained on three other datasets (CIFAR-100, SVHN and Food-101), Stepwise PathNet was up to 8% and 10% more accurate than finely tuned and from-scratch approaches, respectively. Also, some of the selected layers were not supported by the layer functions assumed in PathNet.

Using the pulsed laser deposition technique, we have successfully prepared a‐axis oriented PrBa₂Cu_3−xNb_xO_7−z (PBCNO)(x=0–1.5) films and YBa₂Cu₃O_7−x (YBCO)/PBCNO heterorepitaxial multilayer structures. For the PBCNO films, the resistivity showed semiconducting behavior at all Nb substitution levels investigated. The resistivity increased with increasing Nb content, reaching a value of 10⁴ Ω cm at 100 K for a 300‐Å‐thick film with x=1.0. Using (100)SrTiO₃ (STO) as a substrate, the following multilayer structures were successfully formed: (100)PBCNO(x=1.0)/(100)YBCO/(100)PBCO//STO, (001) YBCO/(100)PBCNO(x=1.0)/(100)PBCO//STO, and (100)YBCO/(100)PBCO/(100)PBCNO(x=1.0)/(100)PBCO//STO. The bottom and top YBCO layers in the (100)YBCO/(100)PBCO/(100)PBCNO/(100)PBCO/(100)YBCO//STO structure showed superconducting transition temperatures (zero resistance) of 82 and 83 K, respectively. The interlayer resistivity of the 200‐Å‐thick PBCNO(x=1) interlayer was as high as 10⁵ Ω cm at 30 K.

Discretization of a regular continuous-time descriptor-system, whose initial condition is consistent with its input, is considered using a general mapping method presented in our previous paper. The proposed mapping discrete-time model is shown to be a proper discretization under the definition explained in the paper. This assures that the response of the mapping model approaches that of the continuous-time descriptor system as the sampling period approaches zero. The consistency of initial conditions for the discrete-time model is also studied and the long-standing issue of ambiguities surrounding irregularities of discrete-time responses at the initial time are clarified with a simple solution. A proper range of design parameters are investigated and their suitable choices suggested. To illustrate the use of the proposed method, a simple circuit that cannot be expressed in the ordinary state-space form is considered. Its responses to a sinusoidal input when started from the consistent and inconsistent initial conditions are simulated to show that the irregularities at the initial time can be overcome easily. The proposed technique provides a convenient simulation and design environment for handling discrete-time systems in a unified manner with consistency and ease.