Abstract: Disclosed are a method for implementing and developing a network model and a related product. The method includes: an establishment command of a neural network model is received, and a computation graph of an initial neural network model is established accordingly; a computing module selected and a connection relationship of the computing module are acquired, the computing module and the connection relationship are added into the computation graph of the initial neural network model, and an intermediate neural network model is obtained; and an establishment ending command of the neural network model is collected, the intermediate neural network model is verified according to the ending command to determine whether the computing module conflicts with other computing modules or not; if not, an established neural network model matched with a computation graph of the intermediate network model is generated, and execution codes matched with the computation graph are generated.

Abstract: The disclosure belongs to the field of microelectronics, and specifically, relates to a method of inducing crystallization of a chalcogenide phase-change material and application thereof. To be specific, a dielectric material is brought into contact with an interface of the chalcogenide phase-change material. The dielectric material is in an octahedral configuration, and the dielectric material provides a crystal nucleus growth center for the crystallization of the chalcogenide phase-change material at the interface between the two, so as to induce the phase-change material to accelerate the crystallization. The method is further applied in a phase-change memory cell. Among all the dielectric material layers in contact with the chalcogenide phase-change material layer, the dielectric material structure of at least one side of the dielectric material layer is an octahedral configuration.

Abstract: Techniques for training neural networks in accordance with an adaptive loss scaling scheme are disclosed. One aspect of the present disclosure relates to a method of training a neural network including a plurality of layers, including determining, by one or more processors, layer-wise loss scale factors for the respective layers and updating, by the one or more processors, parameters for the layers in accordance with error gradients for the layers, wherein the error gradients are scaled with the corresponding layer-wise loss scale factors.

Abstract: Disclosed are a network model compiler and a related product. The network model compiler includes a data IO unit, a compression unit and a storage unit. The data IO unit has one port connected to a data output port of a first computing platform and another port connected to a data input/output port of a second computing platform.

Abstract: Disclosed are a method for implementing and developing a network model and a related product. The method includes: an establishment command of a neural network model is received, and a computation graph of an initial neural network model is established accordingly; a computing module selected and a connection relationship of the computing module are acquired, the computing module and the connection relationship are added into the computation graph of the initial neural network model, and an intermediate neural network model is obtained; and an establishment ending command of the neural network model is collected, the intermediate neural network model is verified according to the ending command to determine whether the computing module conflicts with other computing modules or not; if not, an established neural network model matched with a computation graph of the intermediate network model is generated, and execution codes matched with the computation graph are generated.

Abstract: Disclosed are a method for operation of a network model and a related product. The method includes: a weight data group sent by a network model compiler is received; p-layer weight data of the network model is updated according to the weight data group to obtain an updated network model; and preset data is extracted, the preset data is input as input data into the updated network model for operation to obtain an output result, and the output result is displayed.