神经网络架构搜索研究进展与展望

神经网络架构搜索旨在针对不同任务，自动化地搜索得到性能最优的神经网络结构，是深度学习、计算机视觉技术结合当前现实需求应运而生的一大重要科学问题。对近年来神经网络架构搜索研究进行梳理、归类和评述；阐述神经网络架构搜索的定义和意义，全方位剖析当前研究所面临的难点与挑战；以此为基础，对主流的搜索策略进行阐述和归纳；探讨研究潜在的问题及未来颇具潜力的研究方向，以期推动该领域的进一步发展。

State of the art and prospects of neural architecture search

Considering the distributed intermittent interference threat faced by global navigation satellite system receiver, the anti-jamming performance of STAP(space-time adaptive processing) based on sample matrix inversion method was comprehensively and accurately analyzed. Based on the performance analysis model, according to the mismatch of sampling covariance matrix caused by intermittent interferences, the performance of STAP under different conditions was comprehensively analyzed through classification evaluation, combing induction and theoretical derivation. The analysis results show that when the sampling length is less than the scintillation period and the pre-sample method is used, the missed sampling of space-time adaptive processor will occur and the interference suppression performance will decrease sharply, and the frequency of missed sampling is the sum of the scintillation frequencies of each interference. Numerical and simulation analyses verify the results of analysis, and based on this, the optimal design of STAP in the face of distributed intermittent interference is discussed.