Matches in SemOpenAlex for { <https://semopenalex.org/work/W3173670447> ?p ?o ?g. }
- W3173670447 abstract "Convolutional neural networks (CNNs) have developed to become powerful models for various computer vision tasks ranging from object detection to semantic segmentation. However, most of the state-of-the-art CNNs cannot be deployed directly on edge devices such as smartphones and drones, which need low latency under limited power and memory bandwidth. One popular, straightforward approach to compressing CNNs is network slimming, which imposes $ell_1$ regularization on the channel-associated scaling factors via the batch normalization layers during training. Network slimming thereby identifies insignificant channels that can be pruned for inference. In this paper, we propose replacing the $ell_1$ penalty with an alternative nonconvex, sparsity-inducing penalty in order to yield a more compressed and/or accurate CNN architecture. We investigate $ell_p (0 < p < 1)$, transformed $ell_1$ (T$ell_1$), minimax concave penalty (MCP), and smoothly clipped absolute deviation (SCAD) due to their recent successes and popularity in solving sparse optimization problems, such as compressed sensing and variable selection. We demonstrate the effectiveness of network slimming with nonconvex penalties on three neural network architectures -- VGG-19, DenseNet-40, and ResNet-164 -- on standard image classification datasets. Based on the numerical experiments, T$ell_1$ preserves model accuracy against channel pruning, $ell_{1/2, 3/4}$ yield better compressed models with similar accuracies after retraining as $ell_1$, and MCP and SCAD provide more accurate models after retraining with similar compression as $ell_1$. Network slimming with T$ell_1$ regularization also outperforms the latest Bayesian modification of network slimming in compressing a CNN architecture in terms of memory storage while preserving its model accuracy after channel pruning." @default.
- W3173670447 created "2021-07-05" @default.
- W3173670447 creator A5007322337 @default.
- W3173670447 creator A5016173634 @default.
- W3173670447 creator A5030190007 @default.
- W3173670447 creator A5059708262 @default.
- W3173670447 creator A5079155274 @default.
- W3173670447 date "2020-10-02" @default.
- W3173670447 modified "2023-09-24" @default.
- W3173670447 title "Improving Network Slimming with Nonconvex Regularization" @default.
- W3173670447 cites W104184427 @default.
- W3173670447 cites W1165639838 @default.
- W3173670447 cites W1677182931 @default.
- W3173670447 cites W1686810756 @default.
- W3173670447 cites W1799366690 @default.
- W3173670447 cites W1836465849 @default.
- W3173670447 cites W1901129140 @default.
- W3173670447 cites W1903029394 @default.
- W3173670447 cites W1908384797 @default.
- W3173670447 cites W1943431218 @default.
- W3173670447 cites W1965125844 @default.
- W3173670447 cites W1972163814 @default.
- W3173670447 cites W1991857131 @default.
- W3173670447 cites W1996287810 @default.
- W3173670447 cites W2004544971 @default.
- W3173670447 cites W2014360396 @default.
- W3173670447 cites W2015263936 @default.
- W3173670447 cites W2025666718 @default.
- W3173670447 cites W2032944446 @default.
- W3173670447 cites W2051864124 @default.
- W3173670447 cites W2056201402 @default.
- W3173670447 cites W2074682976 @default.
- W3173670447 cites W2089106993 @default.
- W3173670447 cites W2101675075 @default.
- W3173670447 cites W2102605133 @default.
- W3173670447 cites W2138019504 @default.
- W3173670447 cites W2138049103 @default.
- W3173670447 cites W2145096794 @default.
- W3173670447 cites W2155024610 @default.
- W3173670447 cites W2158928439 @default.
- W3173670447 cites W2163605009 @default.
- W3173670447 cites W2164452299 @default.
- W3173670447 cites W2167215970 @default.
- W3173670447 cites W2168745297 @default.
- W3173670447 cites W2172166488 @default.
- W3173670447 cites W2181101938 @default.
- W3173670447 cites W2183341477 @default.
- W3173670447 cites W2194775991 @default.
- W3173670447 cites W2294059674 @default.
- W3173670447 cites W2331143823 @default.
- W3173670447 cites W2335728318 @default.
- W3173670447 cites W2405920868 @default.
- W3173670447 cites W2412782625 @default.
- W3173670447 cites W2460144244 @default.
- W3173670447 cites W2495425901 @default.
- W3173670447 cites W2515385951 @default.
- W3173670447 cites W2554931888 @default.
- W3173670447 cites W2557728737 @default.
- W3173670447 cites W2560017826 @default.
- W3173670447 cites W2593303827 @default.
- W3173670447 cites W2605135468 @default.
- W3173670447 cites W2613718673 @default.
- W3173670447 cites W2743133559 @default.
- W3173670447 cites W2763531044 @default.
- W3173670447 cites W2795783309 @default.
- W3173670447 cites W2886697509 @default.
- W3173670447 cites W2888333721 @default.
- W3173670447 cites W2896556344 @default.
- W3173670447 cites W2902470604 @default.
- W3173670447 cites W2950967261 @default.
- W3173670447 cites W2962689221 @default.
- W3173670447 cites W2962851801 @default.
- W3173670447 cites W2962958489 @default.
- W3173670447 cites W2962963202 @default.
- W3173670447 cites W2963000224 @default.
- W3173670447 cites W2963114950 @default.
- W3173670447 cites W2963125010 @default.
- W3173670447 cites W2963162885 @default.
- W3173670447 cites W2963163009 @default.
- W3173670447 cites W2963446712 @default.
- W3173670447 cites W2963674932 @default.
- W3173670447 cites W2964010532 @default.
- W3173670447 cites W2964088520 @default.
- W3173670447 cites W2964266063 @default.
- W3173670447 cites W2964461714 @default.
- W3173670447 cites W3008070224 @default.
- W3173670447 cites W3034818206 @default.
- W3173670447 cites W3100817920 @default.
- W3173670447 cites W3106108064 @default.
- W3173670447 cites W3117037513 @default.
- W3173670447 cites W3118608800 @default.
- W3173670447 cites W3119817362 @default.
- W3173670447 cites W3129991176 @default.
- W3173670447 cites W3135817118 @default.
- W3173670447 cites W3138506499 @default.
- W3173670447 cites W3144504312 @default.
- W3173670447 cites W3145472281 @default.
- W3173670447 cites W317954863 @default.
- W3173670447 cites W3182242813 @default.
- W3173670447 cites W591794949 @default.