發(fā)簡信
IP屬地：山西
-
深度學(xué)習(xí)調(diào)參技巧
從粗到細(xì) 實踐中，一般先進(jìn)行初步范圍搜索，然后根據(jù)好結(jié)果出現(xiàn)的地方，再縮小范圍進(jìn)行更精細(xì)的搜索。 先參考相關(guān)論文，以論文中給出的參數(shù)作為初始參數(shù)...
-
Knowledge transfer summary-2
所謂“知識精煉”我的理解就是將一個訓(xùn)練好的復(fù)雜模型的“知識”遷移到一個結(jié)構(gòu)更為簡單的網(wǎng)絡(luò)中，或者通過簡單的網(wǎng)絡(luò)去學(xué)習(xí)復(fù)雜模型中“知識”或模仿復(fù)雜...
-
網(wǎng)絡(luò)量化小結(jié)
網(wǎng)絡(luò)量化作為一種重要的模型壓縮方法，大致可以分為兩類： 直接降低參數(shù)精度典型的工作包括二值網(wǎng)絡(luò)，三值網(wǎng)絡(luò)以及XNOR-Net. HORQ和Net...
-
Speeding-up CNN using CP-Decomposition
Approach We propose a simple two-step approach for speeding up convoluti...
-
Exploiting Linear Structure Within CNNApproach Matrix Decomposition Higher Order Tensor Approximations Monochr...
-
Dynamic Network SurgeryApproach Han song recently propose to compress DNN by deleting unimporta...
-
Speed up CNN with Low Rank ExpansionsApproach Experiment References:Speeding up Convolutional Neural Networks...
-
Learning Structured Sparsity in DNNApproach The optimization target of learning the filter-wise and channel...
-
Fixed-point Factorized NetworksApproach Fixed-point Factorization Full-precision Weights Recovery The q...