Abstract: A method for enhancing kernel reparameterization of a non-linear machine learning model includes providing a predefined machine learning model, expanding a kernel of the predefined machine learning model with a non-linear network for convolution operation of the predefined machine learning model to generate the non-linear machine learning model, training the non-linear machine learning model, reparameterizing the non-linear network back to a kernel for convolution operation of the non-linear machine learning model to generate a reparameterized machine learning model, and deploying the reparameterized machine learning model to an edge device.

Abstract: A method for estimating a flight time of a hydrogen fuel cell UAV (unmanned aerial vehicle) includes multiple steps performed by a controller: obtaining an internal pressure of a hydrogen tank by a pressure sensor installed on the hydrogen tank, calculating a remaining hydrogen volume according to the internal pressure and a capacity of the hydrogen tank, obtaining a reaction current value of the fuel cell, calculating a first hydrogen consumption rate according to the reaction current value, the number of a set of membrane electrodes connected in series and a Faraday constant, obtaining a second hydrogen consumption rate of a purge operation of an anode of the full cell; obtaining a hydrogen leakage rate of a stack of the fuel cell, and calculating the flight time according to the remaining hydrogen volume, the first hydrogen consumption rate, the second hydrogen consumption rate and the hydrogen leakage rate.

Abstract: In aspects of the disclosure, a method, a system, and a computer-readable medium are provided. The method of building a kernel reparameterization for replacing a convolution-wise operation kernel in training of a neural network comprises selecting one or more blocks from tensor blocks and operations; and connecting the selected one or more blocks with the selected operations to build the kernel reparameterization. The kernel reparameterization has a dimension same as that of the convolution-wise operation kernel.

Abstract: In a method of manufacturing a semiconductor device, a fin structure having a channel region protruding from an isolation insulating layer disposed over a semiconductor substrate is formed, a cleaning operation is performed, and an epitaxial semiconductor layer is formed over the channel region. The cleaning operation and the forming the epitaxial semiconductor layer are performed in a same chamber without breaking vacuum.

Abstract: The present invention relates to a protective film for electronic device. The protective film comprises a protective-film body and a first microstructure layer. The protective-film body has a first surface and a second surface. The first microstructure layer is disposed on the first surface of the protective-film body. The second surface of the protective-film body covers the surface of an electronic device. The hardness of the protective film is between 1 H and 12 H. Thanks to it high surface hardness, the protective film according to the present invention has superior wear and scrape resistance. Thereby, the life time of the protective film is extended and hence ensuring protection of the surface of the electronic device.