本文關(guān)鍵詞： 稀疏學(xué)習(xí) 矩陣補全 壓縮感知 矩陣分解 隨機優(yōu)化　出處：《軟件學(xué)報》2017年06期 　論文類型：期刊論文

【摘要】：近年來,隨著壓縮感知技術(shù)在信號處理領(lǐng)域的巨大成功,由其衍生而來的矩陣補全技術(shù)也日益成為機器學(xué)習(xí)領(lǐng)域的研究熱點,諸多研究者針對矩陣補全問題展開了大量卓有成效的研究.為了更好地把握矩陣補全技術(shù)的發(fā)展規(guī)律,促進矩陣補全理論與工程應(yīng)用相結(jié)合,針對矩陣補全模型及其算法進行了綜述.首先,對矩陣補全技術(shù)進行溯源,介紹了從壓縮感知到矩陣補全的自然演化歷程,指出壓縮感知理論的發(fā)展為矩陣補全理論的形成奠定了基礎(chǔ);其次,從非凸非光滑秩函數(shù)松弛的角度將現(xiàn)有矩陣補全模型進行分類,旨在為面向具體應(yīng)用的矩陣補全問題建模提供思路;然后綜述了適用于矩陣補全模型求解的代表性優(yōu)化算法,其目的在于從本質(zhì)上理解各種矩陣補全模型優(yōu)化技巧,從而有利于面向應(yīng)用問題的矩陣補全新模型求解;最后分析了矩陣補全模型及其算法目前存在的問題,提出了可能的解決思路,并對未來的研究方向進行了展望.
[Abstract]:In recent years, with the great success of compressed sensing technology in the field of signal processing, matrix complement technology derived from it has become a hot topic in the field of machine learning. Many researchers have carried out a great deal of fruitful research on the problem of matrix complement, in order to better grasp the development law of matrix complement technology and promote the combination of matrix complement theory and engineering application, This paper summarizes the matrix complement model and its algorithm. Firstly, the source of matrix complement technology is traced, and the natural evolution from compression perception to matrix complement is introduced. It is pointed out that the development of compressed perception theory lays a foundation for the formation of matrix complement theory. Secondly, the existing matrix complement models are classified from the point of view of nonconvex nonsmooth rank function relaxation. The purpose of this paper is to provide ideas for the modeling of matrix complement problems oriented to specific applications, and then summarize the representative optimization algorithms suitable for solving matrix complement models, the purpose of which is to understand various optimization techniques of matrix complement models in essence. Finally, the problems existing in the matrix complement model and its algorithm are analyzed, the possible solutions are put forward, and the future research direction is prospected.
【作者單位】： 南京航空航天大學(xué)計算機科學(xué)與技術(shù)學(xué)院;江蘇省無線傳感網(wǎng)高技術(shù)研究重點實驗室(南京郵電大學(xué));南京郵電大學(xué)計算機學(xué)院;
【基金】：國家自然科學(xué)基金(61472186,61572263,61403208) 江蘇省自然科學(xué)基金(BK20161516,BK20151511) 中國博士后科學(xué)基金(2015M581794) 江蘇省高校自然科學(xué)研究面上項目(15KJB520027) 江蘇省博士后科研資助計劃(1501023C) 南京郵電大學(xué)校級科研基金(NY214127,NY215097)~~
【分類號】：TP181

【相似文獻】

相關(guān)期刊論文 前3條

1 譚瓊;如何形成網(wǎng)絡(luò)流規(guī)劃中的路矩陣[J];系統(tǒng)工程理論與實踐;1992年04期

2 蔣定德;胡光岷;;基于廣義回歸神經(jīng)網(wǎng)絡(luò)的流量矩陣估計[J];計算機應(yīng)用研究;2009年07期

3 ;[J];;年期

相關(guān)碩士學(xué)位論文 前1條

1 唐云;基于Spark的大規(guī)模分布式矩陣運算算法研究與實現(xiàn)[D];南京大學(xué);2016年

，

本文編號：1526790