7 results about "Integer matrix" patented technology

Various examples are directed to systems and methods for hashing an input string. A hashing engine may generate a 1 by N input vector using the input string, where N is a constant number. The hashing engine may expand the input vector to an input matrix, the input matrix having a size of N by N. The hashing engine may access a first integer matrix having a size of N by N and add the first integer matrix and the input matrix to generate an intermediate matrix. The hashing engine may generate a symbol-encoded vector using the intermediate matrix. The hashing engine may also generate a first round output matrix using the symbol-encoded vector and the intermediate matrix and a hash for the input string that is based at least in part on the first round output matrix.

The invention relates to a three-dimensional image encryption method and system based on a four-dimensional hyper-chaotic system and a medium in the technical field of information security. The three-dimensional image encryption method comprises the following steps: mapping all vertexes in V into bytes to obtain an original byte matrix; a four-dimensional hyper-chaotic system is constructed, and a row and column key Kperm and a diffusion key Kdpff are generated. Firstly, scrambling is carried out through the Kperm pair, and then diffusion is carried out through the Kdpff. According to the method, grid vertex coordinates of a three-dimensional image are equivalently mapped into a two-dimensional unsigned integer matrix, on the basis, a designed four-dimensional hyperchaotic system is introduced to respectively generate a row scrambling sequence, a column scrambling sequence and a diffusion mask matrix, and scrambling and XOR diffusion processing is performed on vertex data on a byte level, so that a three-dimensional image is obtained. Therefore, while the encryption security is ensured, the encryption and decryption reversibility is high, and the method can be flexibly expanded under different file formats, different chaotic parameters and different projection strategies, and is suitable for engineering implementation and subsequent optimization.

The invention provides a mass hydrology and water conservancy data compression method and system based on small-scale stationary features, and the method comprises the steps: obtaining hydrology and water conservancy data, and inputting a time vector data set of hydrology and water conservancy of a specific watershed; distinguishing small-scale data and large-scale data of the hydrological and hydraulic data, and performing iterative compression on the small-scale data; calculating an average value and a standard deviation of the small-scale data set, and approximately representing small-scale data by adopting normal distribution; calculating the deviation between each data in the small-scale data set and the average value, approximately representing the deviation by adopting an integer number of resolution standard deviations, and constructing integer matrix storage; and distributing all the small-scale data in a preset range, adopting a plurality of bits to sequentially encode each integer compression data in the integer matrix in a fixed-length manner, and using Huffman to encode the compression data according to different characteristics of different integer distribution frequencies. According to the method, the data with controllable precision loss is constructed to approximately represent the original data, so that the information entropy of the small-scale region in the data stream is effectively reduced.

The invention provides a single-precision floating-point matrix multiplication calculation unit, and the unit comprises a preprocessing module which is used for reading single-precision floating-point data, and dividing the mantissa field part of each matrix into a plurality of groups of integer matrixes with low bit width; the linear evaluation arithmetic logic module is used for executing linear combination operation on the multiple groups of low-bit-width integer matrixes to generate multiple groups of point value matrixes; the low-bit-width integer calculation array is used for executing low-bit-width integer matrix multiplication operation on the multiple groups of point value matrixes to obtain multiple groups of point value product matrixes; the interpolation merging module is used for executing interpolation operation and weighted merging on the multiple groups of point value product matrixes to generate a mantissa field product result; and the format conversion module is used for combining the mantissa field product result with the exponential field part of each matrix to generate a single-precision floating-point number matrix multiplication result. According to the invention, the overall throughput and energy efficiency are improved, and the mapping calculation and system overhead are reduced.

Techniques for training deep neural networks with 4-bit integers are described herein. According to one aspect of the present disclosure, a computer-implemented method for accelerating operations of a deep neural network includes, for each of a plurality of matrix multiplications during forward propagation, obtaining a first matrix by multiplying an activation matrix with a block diagonal Hadamard transform matrix, the activation matrix is an input of the matrix multiplication; obtaining a second matrix by multiplying a weight matrix with the block diagonal Hadamard transform matrix wherein the transpose of the weight matrix is another input of the matrix multiplication; quantizing the first matrix and the second matrix into a 4-bit integer matrix; obtaining a result matrix by multiplying the transpose of the quantized first matrix with the transpose of the quantized second matrix; and inversely quantizing the result matrix into a matrix represented by floating points.

Techniques for training deep neural networks with 4-bit integers are described.According to one aspect of the present disclosure, a computer-implemented method for accelerating operations of a deep neural network comprises: for each of a plurality of matrix multiplications during forward propagation, obtaining a first matrix by multiplying an activation matrix, which is an input to the matrix multiplication, by a block-diagonal Hadamard transform matrix; obtaining a second matrix by multiplying a weighting matrix, one transpose of which is another input to the matrix multiplication, by the block-diagonal Hadamard transform matrix; quantizing the first matrix and the second matrix into 4-bit integer matrices; obtaining a resultant matrix by multiplying the quantized first matrix and the transpose of the quantized second matrix; and dequantizing the resultant matrix into a floating-point matrix.

The invention provides a generative image steganography method and system, computer equipment and a storage medium, and belongs to the field of information hiding, and the method comprises the following steps: an embedding and extraction stage: during embedding, generating a pseudo-random matrix by using a shared key, dividing [0, 1) intervals according to an embedding rate R by combining a secret message matrix, generating an epipolar probability matrix, and extracting the pseudo-random matrix from the epipolar probability matrix; mapping into a standard normal distribution noise matrix through an inverse cumulative distribution function, and inputting the standard normal distribution noise matrix into a pre-training diffusion model to obtain a steganographic image; during extraction, a receiver synchronously generates a reference image, obtains initial message estimation through a pixel Euclidean distance, converts the initial message estimation into a continuous matrix, generates an estimation image through a diffusion model, iteratively optimizes a loss distance in combination with gradient descent, and converts the initial message estimation into an integer matrix until the loss reaches the standard to recover a secret message. According to the method, a diffusion model does not need to be modified, the unification of high embedding capacity, nearly 99% extraction accuracy and high safety is realized, and the method is adaptive to diversified scenes.