3 results about "Toeplitz matrix" patented technology

The application provides a language model parameter efficient fine-tuning method and device based on a block Toeplitz matrix, can realize high-rank, dense weight update and is suitable for an efficient parameter fine-tuning method of a large-scale pre-training model, so as to improve the adaptability and overall performance of the model in various downstream tasks. In order to realize the above technical effects, the core technical point of the application is to provide a parameter efficient fine-tuning method based on a block Toeplitz structure, by introducing multiple Toeplitz structure sub-blocks in the weight update of the pre-training model, the weight update space is structured. The method can generate a high-rank, dense weight update matrix under the condition that the number of parameters is controlled, so as to significantly improve the expression ability of the fine-tuning method.

The application provides a method for constructing a reconfigurable sparse linear array, which comprises the following steps: firstly, sampling a target beam to obtain a matrix constituted by sampling beams; secondly, obtaining a rank-minimum Toeplitz matrix by a weighted atom norm minimization method; thirdly, estimating the frequency and weight of the atom by using a Root-MUSIC algorithm; and finally, converting the frequency and weight into the element position and excitation of the reconfigurable sparse linear array through a mapping relationship. The RANM algorithm provided by the application has a performance advantage in the sparsity, can reduce the number of elements under the condition that the shape of the radiation beam pattern is almost unchanged, and thus reduces the complexity and power consumption of the system. The algorithm avoids the grid mismatch problem existing in the traditional sparse recovery algorithm, and thus is superior to the traditional reconfigurable sparse linear array algorithm in the matching accuracy of the reconstructed beam.

This invention discloses a super-resolution forward-looking imaging method for real-aperture scanning radar, relating to the field of radar technology. The method includes: establishing an echo signal model in the forward-looking imaging mode of a real-aperture scanning radar; performing pulse compression and range migration correction on the target echo signal obtained from the echo signal model to obtain a pre-processed target echo signal; correcting beam distortion of the existing echo signal model using a block approximation method, transforming the spatially variable convolution matrix into a locally spatially invariant Toeplitz matrix; and combining the beam-distortion-corrected echo signal model with the Gaussian maximum a posteriori algorithm and introducing the Barzilai-Borwein method to adaptively adjust the iteration step size to achieve super-resolution imaging. This solves the problem of insufficient azimuth resolution in forward-looking imaging and improves the signal-to-noise ratio of the imaging results.