3 results about "Zero filling" patented technology

Disclosed is a method that includes generating by an output processing node of a first section of a processing graph, a plurality of output tiles of an output tensor. The plurality of output tiles of the output tensor is written in a memory, where the writing includes zero-padding the plurality of output tiles of the output tensor in the memory. The zero-padded plurality of output tiles of the output tensor are tiled, to generate a plurality of input tiles of an input tensor. The plurality of input tiles of the input tensor is processed in a second section of the processing graph.

The invention relates to the technical field of computer information security, and discloses an AI data security encryption method and system for an electronic file. An electronic file is read byte by byte and converted into a binary bit stream, a main matrix is filled according to a preset row and column sequence, and insufficient parts are subjected to zero filling to ensure uniform structure. Receiving a master key seed, generating a sub-key parameter corresponding to the matrix scale according to a derivation rule, and generating two sets of key matrixes element by element according to the sub-key parameter; dividing the main matrix into two parts which are not overlapped with each other according to the parity of the row and column index sum, performing XOR on the two parts and the corresponding key matrix to form a double-track encryption matrix, and fusing to obtain a main encryption matrix. And then, column cyclic displacement disturbance is driven by using sub-key parameters, and an in-column structure is disrupted. And finally, carrying out row-based preferential expansion and stripping zero padding, and outputting a compact ciphertext. The method covers a complete process from file transcoding, key generation to double-track encryption, fusion disturbance and serialization output.

The present application belongs to the technical field of lossless compression of time series data, and provides an adaptive lossless compression method for oscillating floating point data storage. In order to meet the needs of high efficiency and low delay of time series data processing, based on the bit distribution analysis of the exclusive or result of adjacent data points, the dynamic detection and compact coding strategy can effectively mine the structural redundancy such as high bit prefix and low bit zero padding, realize efficient compression and fast operation. The algorithm can ensure the lossless recovery of data, improve the compression ratio and the decompression speed. Secondly, the present application integrates branch judgment and adaptive coding mechanism, which can flexibly select the coding path according to the characteristics of different data segments. Through the mechanisms of prefix bit, minimal branch and high bit jump, the present application improves the adaptability to periodic structure, long-range correlation and data distribution change, and further optimizes the compression efficiency.