Abstract: An acceleration and compression method for a deep convolutional neural network based on quantization of a parameter provided by the present application comprises: quantizing the parameter of the deep convolutional neural network to obtain a plurality of subcode books and respective corresponding index values of the plurality of subcode books; acquiring an output feature map of the deep convolutional neural network according to the plurality of subcode books and respective corresponding index values of the plurality of subcode books. The present application may implement the acceleration and compression for a deep convolutional neural network.

Abstract: A method including receiving, by a processor, a computing instruction for a neural network, wherein the computing instruction for the neural network includes a computing rule for the neural network and a connection weight of the neural network, and the connection weight is a power of 2; and inputting, for a multiplication operation in the computing rule for the neural network, a source operand corresponding to the multiplication operation to a shift register, and performing a shift operation based on a connection weight corresponding to the multiplication operation, wherein the shift register outputs a target result operand as a result of the multiplication operation. The neural network uses a shift operation, and a neural network computing speed is increased.

Abstract: A method including receiving, by a processor, a computing instruction for a neural network, wherein the computing instruction for the neural network includes a computing rule for the neural network and a connection weight of the neural network, and the connection weight is a power of 2; and inputting, for a multiplication operation in the computing rule for the neural network, a source operand corresponding to the multiplication operation to a shift register, and performing a shift operation based on a connection weight corresponding to the multiplication operation, wherein the shift register outputs a target result operand as a result of the multiplication operation. The neural network uses a shift operation, and a neural network computing speed is increased.

Abstract: A method, an apparatus, a device and a computer readable storage media for neural network processing are disclosed. The method includes constructing a training function having a constraint for a neural network; and solving a constrained optimization based on the training function to obtain connection weights of the neural network. The neural network processing method, apparatus, device and computer readable storage media of the embodiments of the present disclosure model a problem of solving connection weights of a neural network from the perspective of an optimization problem, and can effectively find a solution for the problem of solving the connection weights of the neural network, thus being able to improve the speed of training of the neural network.

Abstract: An acceleration and compression method for a deep convolutional neural network based on quantization of a parameter provided by the present application comprises: quantizing the parameter of the deep convolutional neural network to obtain a plurality of subcode books and respective corresponding index values of the plurality of subcode books; acquiring an output feature map of the deep convolutional neural network according to the plurality of subcode books and respective corresponding index values of the plurality of subcode books. The present application may implement the acceleration and compression for a deep convolutional neural network.

Abstract: An image matching method based on cascaded binary encoding includes using a hashing look-up with multiple hashing tables to coarsely filter candidate key-points in an image to produce a candidate subset of key-points, projecting the candidate subset into a high-dimensional Hamming space, and building a “Hamming distance-memory address” hashing table. An optimal matching key-point is discovered by querying the hashing table. The image matching method has high processing speed and matching quality, which can be used for efficient and accurate image matching.

Abstract: This invention involves an image matching method based on the cascaded binary encoding. The stated method includes: Procedure S1, using the hashing look-up with multiple hashing tables to coarsely filter candidate key-points in the image to produce a candidate subset of key-points; Procedure S2, projecting the candidate subset into a high-dimensional Hamming space; Procedure S3, a “Hamming distance-memory address” hashing table is built, and the optimal matching key-point is discovered by querying this hashing table. The image matching method proposed in this invention has high processing speed and matching quality, which can be used for efficient and accurate image matching.