Maintained by Difan Deng and Marius Lindauer.
The following list considers papers related to neural architecture search. It is by no means complete. If you miss a paper on the list, please let us know.
Please note that although NAS methods steadily improve, the quality of empirical evaluations in this field are still lagging behind compared to other areas in machine learning, AI and optimization. We would therefore like to share some best practices for empirical evaluations of NAS methods, which we believe will facilitate sustained and measurable progress in the field. If you are interested in a teaser, please read our blog post or directly jump to our checklist.
2021 |
|
| 1147. | Tsokov, Stefan ; Lazarova, Milena ; Aleksieva-Petrova, Adelina An Evolutionary Approach to the Design of Convolutional Neural Networks for Human Activity Recognition Journal Article Indian Journal of Computer Science and Engineering, 2021. @article{Tsokov2021, title = {An Evolutionary Approach to the Design of Convolutional Neural Networks for Human Activity Recognition}, author = {Tsokov, Stefan and Lazarova, Milena and Aleksieva-Petrova, Adelina}, url = {http://www.ijcse.com/docs/INDJCSE21-12-02-145.pdf}, year = {2021}, date = {2021-06-01}, journal = {Indian Journal of Computer Science and Engineering}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1146. | Chen, Wuyang; Gong, Xinyu; Wang, Zhangyang Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective Inproceedings ICLR 2021, 2021. @inproceedings{chen2021neural, title = {Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective}, author = {Wuyang Chen and Xinyu Gong and Zhangyang Wang}, url = {https://arxiv.org/abs/2102.11535}, year = {2021}, date = {2021-05-04}, booktitle = {ICLR 2021}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1145. | Zhang, Yi ; Liu, Yang ; Liu, Shirley X Neural network architecture search with AMBER Journal Article Nature Machine Intelligence, pp. 372-373, 2021. @article{Zhang2021, title = {Neural network architecture search with AMBER}, author = {Zhang, Yi and Liu, Yang and Liu, X. Shirley}, url = {https://doi.org/10.1038/s42256-021-00350-x}, year = {2021}, date = {2021-05-01}, journal = {Nature Machine Intelligence}, pages = {372-373}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1144. | Schoenherr, Georg P The Nonlinearity Coefficient - A Practical Guide to Neural Architecture Design PhD Thesis 2021. @phdthesis{Schoenherr2021, title = {The Nonlinearity Coefficient - A Practical Guide to Neural Architecture Design}, author = {Georg P. Schoenherr}, url = {http://reports-archive.adm.cs.cmu.edu/anon/anon/home/ftp/usr/ftp/2021/abstracts/21-110.html}, year = {2021}, date = {2021-05-01}, keywords = {}, pubstate = {published}, tppubtype = {phdthesis} } |
| 1143. | Yang, Yuxuan; Gao, Zhongke; Li, Yanli; Wang, He A CNN identified by reinforcement learning-based optimization framework for EEG-based state evaluation Journal Article Journal of Neural Engineering, 18 (4), pp. 046059, 2021. @article{Yang_2021, title = {A CNN identified by reinforcement learning-based optimization framework for EEG-based state evaluation}, author = {Yuxuan Yang and Zhongke Gao and Yanli Li and He Wang}, url = {https://doi.org/10.1088/1741-2552/abfa71}, doi = {10.1088/1741-2552/abfa71}, year = {2021}, date = {2021-05-01}, journal = {Journal of Neural Engineering}, volume = {18}, number = {4}, pages = {046059}, publisher = {IOP Publishing}, abstract = {Objective. Electroencephalogram (EEG) data, as a kind of complex time-series, is one of the most widely-used information measurements for evaluating human psychophysiological states. Recently, numerous works applied deep learning techniques, especially the convolutional neural network (CNN), into EEG-based research. The design of the hyper-parameters of the CNN model has a great influence on the performance of the model. Therefore, automatically designing these hyper-parameters can save the time and labor of experts. This leads to the appearance of the neural architecture search technique. In this paper, we propose a reinforcement learning (RL)-based step-by-step framework to efficiently search for CNN models. Approach. Specifically, the deep Q network in RL is first used to determine the depth of convolutional layers and the connection modes among layers. Then particle swarm optimization algorithm is used to fine-tune the number and size of convolution kernels. Through this step-by-step strategy, the search space can be narrowed in each step for saving the overall time cost. This framework is employed for both EEG-based sleep stage classification and driver drowsiness evaluation tasks. Main results. The results show that compared with state-of-the-art methods, the high-performance CNN models identified by the proposed optimization framework, can achieve high overall accuracy and better root mean squared error in the two tasks. Significance. Therefore, the proposed optimization framework has a great potential to provide high-performance results for other kinds of classification and prediction tasks. In this way, it can greatly save researchers’ time cost and promote broader applications of CNNs.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Objective. Electroencephalogram (EEG) data, as a kind of complex time-series, is one of the most widely-used information measurements for evaluating human psychophysiological states. Recently, numerous works applied deep learning techniques, especially the convolutional neural network (CNN), into EEG-based research. The design of the hyper-parameters of the CNN model has a great influence on the performance of the model. Therefore, automatically designing these hyper-parameters can save the time and labor of experts. This leads to the appearance of the neural architecture search technique. In this paper, we propose a reinforcement learning (RL)-based step-by-step framework to efficiently search for CNN models. Approach. Specifically, the deep Q network in RL is first used to determine the depth of convolutional layers and the connection modes among layers. Then particle swarm optimization algorithm is used to fine-tune the number and size of convolution kernels. Through this step-by-step strategy, the search space can be narrowed in each step for saving the overall time cost. This framework is employed for both EEG-based sleep stage classification and driver drowsiness evaluation tasks. Main results. The results show that compared with state-of-the-art methods, the high-performance CNN models identified by the proposed optimization framework, can achieve high overall accuracy and better root mean squared error in the two tasks. Significance. Therefore, the proposed optimization framework has a great potential to provide high-performance results for other kinds of classification and prediction tasks. In this way, it can greatly save researchers’ time cost and promote broader applications of CNNs. |
| 1142. | Pan, Zheyi; Ke, Songyu; Yang, Xiaodu; Liang, Yuxuan; Yu, Yong; Zhang, Junbo; Zheng, Yu AutoSTG: Neural Architecture Search forPredictions of Spatio-Temporal Graphs Inproceedings WWW 2021, 2021. @inproceedings{PanWWW2021, title = {AutoSTG: Neural Architecture Search forPredictions of Spatio-Temporal Graphs}, author = {Zheyi Pan and Songyu Ke and Xiaodu Yang and Yuxuan Liang and Yong Yu and Junbo Zhang and Yu Zheng}, url = {http://panzheyi.cc/publication/pan2021autostg/paper.pdf}, year = {2021}, date = {2021-04-19}, booktitle = {WWW 2021}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1141. | Kyriakides, George ; Margaritis, Konstantinos Evolving graph convolutional networks for neural architecture search Journal Article Neural Computing and Applications, 2021. @article{Kriakides2021Evolving, title = {Evolving graph convolutional networks for neural architecture search}, author = {Kyriakides, George and Margaritis, Konstantinos}, url = {https://doi.org/10.1007/s00521-021-05979-8}, doi = {10.1007/s00521-021-05979-8}, year = {2021}, date = {2021-04-15}, journal = {Neural Computing and Applications}, abstract = {As neural architecture search (NAS) becomes an increasingly adopted method to design network architectures, various methods have been proposed to speedup the process. Besides proxy evaluation tasks, weight sharing, and scaling down the evaluated architectures, performance-predicting models exhibit multiple advantages. Eliminating the need to train candidate architectures and enabling transfer learning between datasets, researchers can also utilize them as a surrogate function for Bayesian optimization. On the other hand, graph convolutional networks (GCNs) have also been increasingly adopted for various tasks, enabling deep learning techniques on graphs without feature engineering. In this paper, we employ an evolutionary-based NAS method to evolve GCNs for the problem of predicting the relative performance of various architectures included in the NAS-Bench-101 dataset. By fine-tuning the architecture generated by our methodology, we manage to achieve a Kendall’s tau correlation coefficient of 0.907 between 1050 completely unseen architectures, utilizing only 450 samples, while also outperforming a strong baseline on the same task. Furthermore, we validate our method on custom global search space architectures, generated for the Fashion-MNIST dataset.}, keywords = {}, pubstate = {published}, tppubtype = {article} } As neural architecture search (NAS) becomes an increasingly adopted method to design network architectures, various methods have been proposed to speedup the process. Besides proxy evaluation tasks, weight sharing, and scaling down the evaluated architectures, performance-predicting models exhibit multiple advantages. Eliminating the need to train candidate architectures and enabling transfer learning between datasets, researchers can also utilize them as a surrogate function for Bayesian optimization. On the other hand, graph convolutional networks (GCNs) have also been increasingly adopted for various tasks, enabling deep learning techniques on graphs without feature engineering. In this paper, we employ an evolutionary-based NAS method to evolve GCNs for the problem of predicting the relative performance of various architectures included in the NAS-Bench-101 dataset. By fine-tuning the architecture generated by our methodology, we manage to achieve a Kendall’s tau correlation coefficient of 0.907 between 1050 completely unseen architectures, utilizing only 450 samples, while also outperforming a strong baseline on the same task. Furthermore, we validate our method on custom global search space architectures, generated for the Fashion-MNIST dataset. |
| 1140. | Wang, Linnan; Xie, Saining; Li, Teng; Fonseca, Rodrigo; Tian, Yuandong Sample-Efficient Neural Architecture Search by Learning Actions for Monte Carlo Tree Search Journal Article IEEE transactions on pattern analysis and machine intelligence, PP , 2021, ISSN: 0162-8828. @article{PMID:33826511, title = {Sample-Efficient Neural Architecture Search by Learning Actions for Monte Carlo Tree Search}, author = {Linnan Wang and Saining Xie and Teng Li and Rodrigo Fonseca and Yuandong Tian}, url = {https://doi.org/10.1109/TPAMI.2021.3071343}, doi = {10.1109/tpami.2021.3071343}, issn = {0162-8828}, year = {2021}, date = {2021-04-01}, journal = {IEEE transactions on pattern analysis and machine intelligence}, volume = {PP}, abstract = {Neural Architecture Search (NAS) has emerged as a promising technique for automatic neural network design. However, existing MCTS based NAS approaches often utilize manually designed action space, which is not directly related to the performance metric to be optimized (e.g., accuracy), leading to sample-inefficient explorations of architectures. To improve the sample efficiency, this paper proposes Latent Action Neural Architecture Search (LaNAS), which learns actions to recursively partition the search space into good or bad regions that contain networks with similar performance metrics. During the search phase, as different action sequences lead to regions with different performance, the search efficiency can be significantly improved by biasing towards the good regions. On three NAS tasks, empirical results demonstrate that LaNAS is at least an order more sample efficient than baseline methods including evolutionary algorithms, Bayesian optimizations, and random search. When applied in practice, both one-shot and regular LaNAS consistently outperform existing results. Particularly, LaNAS achieves 99.0% accuracy on CIFAR-10 and 80.8% top1 accuracy at 600 MFLOPS on ImageNet in only 800 samples, significantly outperforming AmoebaNet with 33x fewer samples. Our code is publicly available at https://github.com/facebookresearch/LaMCTS.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Neural Architecture Search (NAS) has emerged as a promising technique for automatic neural network design. However, existing MCTS based NAS approaches often utilize manually designed action space, which is not directly related to the performance metric to be optimized (e.g., accuracy), leading to sample-inefficient explorations of architectures. To improve the sample efficiency, this paper proposes Latent Action Neural Architecture Search (LaNAS), which learns actions to recursively partition the search space into good or bad regions that contain networks with similar performance metrics. During the search phase, as different action sequences lead to regions with different performance, the search efficiency can be significantly improved by biasing towards the good regions. On three NAS tasks, empirical results demonstrate that LaNAS is at least an order more sample efficient than baseline methods including evolutionary algorithms, Bayesian optimizations, and random search. When applied in practice, both one-shot and regular LaNAS consistently outperform existing results. Particularly, LaNAS achieves 99.0% accuracy on CIFAR-10 and 80.8% top1 accuracy at 600 MFLOPS on ImageNet in only 800 samples, significantly outperforming AmoebaNet with 33x fewer samples. Our code is publicly available at https://github.com/facebookresearch/LaMCTS. |
| 1139. | Zhang, Zhentong ; Shan, Yugang; Yuan, Jie Multi-Level Cell Progressive Differentiable Architecture Search to Improve Image Classification Accuracy Journal Article Journal of Signal Processing Systems, 2021. @article{ZhangJSPS2021, title = {Multi-Level Cell Progressive Differentiable Architecture Search to Improve Image Classification Accuracy}, author = {Zhang, Zhentong and Yugang Shan and Jie Yuan}, url = {https://doi.org/10.1007/s11265-021-01647-1}, year = {2021}, date = {2021-03-08}, journal = {Journal of Signal Processing Systems}, abstract = {In recent years, the neural architecture search has continuously made significant progress in the field of image recognition. Among them, the differentiable method has obvious advantages compared with other search methods in terms of computational cost and accuracy to deal with image classification. However, the differentiable method is usually composed of single cell, which cannot efficiently extract the features of the network. In response to this problem, we propose a multi-level cell progressive differentiable method which allows cells to have different types according to the levels of the network. In differentiable method, the gap between the search network and the evaluation one is large, and the correlation is low. We design an algorithm to improve the distribution of architecture parameters. We also optimize the loss function and use the regularization method of additional action to improve deep network performance. The method achieves good search and classification results on CIFAR10 and ImageNet (mobile setting).}, keywords = {}, pubstate = {published}, tppubtype = {article} } In recent years, the neural architecture search has continuously made significant progress in the field of image recognition. Among them, the differentiable method has obvious advantages compared with other search methods in terms of computational cost and accuracy to deal with image classification. However, the differentiable method is usually composed of single cell, which cannot efficiently extract the features of the network. In response to this problem, we propose a multi-level cell progressive differentiable method which allows cells to have different types according to the levels of the network. In differentiable method, the gap between the search network and the evaluation one is large, and the correlation is low. We design an algorithm to improve the distribution of architecture parameters. We also optimize the loss function and use the regularization method of additional action to improve deep network performance. The method achieves good search and classification results on CIFAR10 and ImageNet (mobile setting). |
| 1138. | Zimmer, Lucas; Lindauer, Marius; Hutter, Frank Auto-Pytorch: Multi-Fidelity MetaLearning for Efficient and Robust AutoDL Journal Article IEEE transactions on pattern analysis and machine intelligence, PP , 2021, ISSN: 0162-8828. @article{PMID:33750687, title = {Auto-Pytorch: Multi-Fidelity MetaLearning for Efficient and Robust AutoDL}, author = {Lucas Zimmer and Marius Lindauer and Frank Hutter}, url = {https://doi.org/10.1109/TPAMI.2021.3067763}, doi = {10.1109/tpami.2021.3067763}, issn = {0162-8828}, year = {2021}, date = {2021-03-01}, journal = {IEEE transactions on pattern analysis and machine intelligence}, volume = {PP}, abstract = {While early AutoML frameworks focused on optimizing traditional ML pipelines and their hyperparameters, a recent trend in AutoML is to focus on neural architecture search. In this paper, we introduce Auto-PyTorch, which brings the best of these two worlds together by jointly and robustly optimizing the architecture of networks and the training hyperparameters to enable fully automated deep learning (AutoDL). Auto-PyTorch achieves state-of-the-art performance on several tabular benchmarks by combining multi-fidelity optimization with portfolio construction for warmstarting and ensembling of deep neural networks (DNNs) and common baselines for tabular data. To thoroughly study our assumptions on how to design such an AutoDL system, we additionally introduce a new benchmark on learning curves for DNNs, dubbed LCBench, and run extensive ablation studies of the full Auto-PyTorch on typical AutoML benchmarks, eventually showing that Auto-PyTorch performs better than several state-of-the-art competitors on average.}, keywords = {}, pubstate = {published}, tppubtype = {article} } While early AutoML frameworks focused on optimizing traditional ML pipelines and their hyperparameters, a recent trend in AutoML is to focus on neural architecture search. In this paper, we introduce Auto-PyTorch, which brings the best of these two worlds together by jointly and robustly optimizing the architecture of networks and the training hyperparameters to enable fully automated deep learning (AutoDL). Auto-PyTorch achieves state-of-the-art performance on several tabular benchmarks by combining multi-fidelity optimization with portfolio construction for warmstarting and ensembling of deep neural networks (DNNs) and common baselines for tabular data. To thoroughly study our assumptions on how to design such an AutoDL system, we additionally introduce a new benchmark on learning curves for DNNs, dubbed LCBench, and run extensive ablation studies of the full Auto-PyTorch on typical AutoML benchmarks, eventually showing that Auto-PyTorch performs better than several state-of-the-art competitors on average. |
| 1137. | Zheng, X; Ji, R; Chen, Y; Wang, Q; Zhang, B; Ye, Q; Chen, J; Huang, F; Tian, Y MIGO-NAS: Towards Fast and Generalizable Neural Architecture Search Journal Article IEEE Transactions on Pattern Analysis & Machine Intelligence, (01), pp. 1-1, 2021, ISSN: 1939-3539. @article{9377468, title = {MIGO-NAS: Towards Fast and Generalizable Neural Architecture Search}, author = {X Zheng and R Ji and Y Chen and Q Wang and B Zhang and Q Ye and J Chen and F Huang and Y Tian}, url = {https://www.computer.org/csdl/journal/tp/5555/01/09377468/1rUNdbz4LQY}, doi = {10.1109/TPAMI.2021.3065138}, issn = {1939-3539}, year = {2021}, date = {2021-03-01}, journal = {IEEE Transactions on Pattern Analysis & Machine Intelligence}, number = {01}, pages = {1-1}, publisher = {IEEE Computer Society}, address = {Los Alamitos, CA, USA}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1136. | Liu, Lanlan; Zhang, Yuting; Deng, Jia; Soatto, Stefano Dynamically Grown Generative Adversarial Networks Inproceedings AAAI 2021, 2021. @inproceedings{LiuAAAI2021, title = {Dynamically Grown Generative Adversarial Networks}, author = {Lanlan Liu and Yuting Zhang and Jia Deng and Stefano Soatto}, url = {https://www.aaai.org/AAAI21Papers/AAAI-1376.LiuL.pdf}, year = {2021}, date = {2021-02-02}, booktitle = {AAAI 2021}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1135. | Xu, Y; Xie, L; Dai, W; Zhang, X; Chen, X; Qi, G; Xiong, H; Tian, Q Partially-Connected Neural Architecture Search for Reduced Computational Redundancy Journal Article IEEE Transactions on Pattern Analysis & Machine Intelligence, (01), pp. 1-1, 2021, ISSN: 1939-3539. @article{9354953, title = {Partially-Connected Neural Architecture Search for Reduced Computational Redundancy}, author = {Y Xu and L Xie and W Dai and X Zhang and X Chen and G Qi and H Xiong and Q Tian}, url = {https://www.computer.org/csdl/journal/tp/5555/01/09354953/1rgCccYlOaQ}, doi = {10.1109/TPAMI.2021.3059510}, issn = {1939-3539}, year = {2021}, date = {2021-02-01}, journal = {IEEE Transactions on Pattern Analysis & Machine Intelligence}, number = {01}, pages = {1-1}, publisher = {IEEE Computer Society}, address = {Los Alamitos, CA, USA}, abstract = {Differentiable architecture search (DARTS) enables effective neural architecture search (NAS) using gradient descent, but suffers from high memory and computational costs. In this paper, we propose a novel approach, namely Partially-Connected DARTS (PC-DARTS), to achieve efficient and stable neural architecture search by reducing the channel and spatial redundancies of the super-network. In the channel level, partial channel connection is presented to randomly sample a small subset of channels for operation selection to accelerate the search process and suppress the over-fitting of the super-network. Side operation is introduced for bypassing (non-sampled) channels to guarantee the performance of searched architectures under extremely low sampling rates. In the spatial level, input features are down-sampled to eliminate spatial redundancy and enhance the efficiency of the mixed computation for operation selection. Furthermore, edge normalization is developed to maintain the consistency of edge selection based on channel sampling with the architectural parameters for edges. Experimental results demonstrate that the proposed approach achieves higher search speed and training stability than DARTS. PC-DARTS obtains a top-1 error rate of 2.55% on CIFAR-10 with 0.07 GPU-days for architecture search, and a state-of-the-art top-1 error rate of 24.1% on ImageNet (under the mobile setting) within 2.8 GPU-day.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Differentiable architecture search (DARTS) enables effective neural architecture search (NAS) using gradient descent, but suffers from high memory and computational costs. In this paper, we propose a novel approach, namely Partially-Connected DARTS (PC-DARTS), to achieve efficient and stable neural architecture search by reducing the channel and spatial redundancies of the super-network. In the channel level, partial channel connection is presented to randomly sample a small subset of channels for operation selection to accelerate the search process and suppress the over-fitting of the super-network. Side operation is introduced for bypassing (non-sampled) channels to guarantee the performance of searched architectures under extremely low sampling rates. In the spatial level, input features are down-sampled to eliminate spatial redundancy and enhance the efficiency of the mixed computation for operation selection. Furthermore, edge normalization is developed to maintain the consistency of edge selection based on channel sampling with the architectural parameters for edges. Experimental results demonstrate that the proposed approach achieves higher search speed and training stability than DARTS. PC-DARTS obtains a top-1 error rate of 2.55% on CIFAR-10 with 0.07 GPU-days for architecture search, and a state-of-the-art top-1 error rate of 24.1% on ImageNet (under the mobile setting) within 2.8 GPU-day. |
| 1134. | Gao, Yanjie; Zhu, Yonghao; Zhang, Hongyu; Lin, Haoxiang; Yang, Mao Resource-Guided Configuration Space Reduction for Deep Learning Models Inproceedings 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE), pp. 175-187, 2021. @inproceedings{9402095, title = {Resource-Guided Configuration Space Reduction for Deep Learning Models}, author = {Yanjie Gao and Yonghao Zhu and Hongyu Zhang and Haoxiang Lin and Mao Yang}, url = {https://ieeexplore.ieee.org/abstract/document/9402095}, doi = {10.1109/ICSE43902.2021.00028}, year = {2021}, date = {2021-01-01}, booktitle = {2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE)}, pages = {175-187}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1133. | Liu, Mengyu; Yin, Hujun Efficient pyramid context encoding and feature embedding for semantic segmentation Journal Article Image and Vision Computing, 111 , pp. 104195, 2021, ISSN: 0262-8856. @article{LIU2021104195, title = {Efficient pyramid context encoding and feature embedding for semantic segmentation}, author = {Mengyu Liu and Hujun Yin}, url = {https://www.sciencedirect.com/science/article/pii/S0262885621001001}, doi = {https://doi.org/10.1016/j.imavis.2021.104195}, issn = {0262-8856}, year = {2021}, date = {2021-01-01}, journal = {Image and Vision Computing}, volume = {111}, pages = {104195}, abstract = {For reality applications of semantic segmentation, inference speed and memory usage are two important factors. To address these challenges, we propose a lightweight feature pyramid encoding network (FPENet) for semantic segmentation with a good trade-off between accuracy and speed. We use a series of feature pyramid encoding (FPE) blocks to encode context at multiple scales in the encoder. Each FPE block consists of different depthwise dilated convolutions that perform as a spatial pyramid to extract features and reduce computational costs. During training, a one-shot neural architecture search algorithm is adopted to find the optimal structure for each FPE block from a large search space with a small search cost. After the search for the encoder, a mutual embedding upsample module is introduced in the decoder, consisting of two attention blocks. The encoder-decoder attention mechanism is used to help aggregate efficiently high-level semantic features and low-level spatial details. The proposed network outperforms the existing real-time methods with fewer parameters and improved inference speed on the Cityscapes and CamVid benchmark datasets. Specifically, it achieved 72.3% mean IoU on the Cityscapes test set with only 0.4 M parameters and 192.6 FPS speed on an Nvidia Titan V100 GPU, and 73.4% mean IoU with 116.2 FPS when running on higher resolution images.}, keywords = {}, pubstate = {published}, tppubtype = {article} } For reality applications of semantic segmentation, inference speed and memory usage are two important factors. To address these challenges, we propose a lightweight feature pyramid encoding network (FPENet) for semantic segmentation with a good trade-off between accuracy and speed. We use a series of feature pyramid encoding (FPE) blocks to encode context at multiple scales in the encoder. Each FPE block consists of different depthwise dilated convolutions that perform as a spatial pyramid to extract features and reduce computational costs. During training, a one-shot neural architecture search algorithm is adopted to find the optimal structure for each FPE block from a large search space with a small search cost. After the search for the encoder, a mutual embedding upsample module is introduced in the decoder, consisting of two attention blocks. The encoder-decoder attention mechanism is used to help aggregate efficiently high-level semantic features and low-level spatial details. The proposed network outperforms the existing real-time methods with fewer parameters and improved inference speed on the Cityscapes and CamVid benchmark datasets. Specifically, it achieved 72.3% mean IoU on the Cityscapes test set with only 0.4 M parameters and 192.6 FPS speed on an Nvidia Titan V100 GPU, and 73.4% mean IoU with 116.2 FPS when running on higher resolution images. |
| 1132. | Chen, Shumei; Yu, Jianbo; Wang, Shijin ISA Transactions, 2021, ISSN: 0019-0578. @article{CHEN2021, title = {One-dimensional convolutional neural network-based active feature extraction for fault detection and diagnosis of industrial processes and its understanding via visualization}, author = {Shumei Chen and Jianbo Yu and Shijin Wang}, url = {https://www.sciencedirect.com/science/article/pii/S0019057821002391}, doi = {https://doi.org/10.1016/j.isatra.2021.04.042}, issn = {0019-0578}, year = {2021}, date = {2021-01-01}, journal = {ISA Transactions}, abstract = {Feature extraction from process signals enables process monitoring models to be effective in industrial processes. Deep learning presents extensive possibilities for extracting abstract features from image and visual data. However, the main inputs of conventional deep neural networks are large images. To overcome this, a one-dimension convolution neural network-based model optimized by a reinforcement-learning-based neural architecture search, is proposed for multivariate processes control. The experimental results illustrate its predominance for detecting and recognizing process faults. Feature and network visualization are also implemented to explore the reasons for its outstanding performance. This research extends the applications of convolutional neural network based on one-dimension process signals in complex multivariate process control.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Feature extraction from process signals enables process monitoring models to be effective in industrial processes. Deep learning presents extensive possibilities for extracting abstract features from image and visual data. However, the main inputs of conventional deep neural networks are large images. To overcome this, a one-dimension convolution neural network-based model optimized by a reinforcement-learning-based neural architecture search, is proposed for multivariate processes control. The experimental results illustrate its predominance for detecting and recognizing process faults. Feature and network visualization are also implemented to explore the reasons for its outstanding performance. This research extends the applications of convolutional neural network based on one-dimension process signals in complex multivariate process control. |
| 1131. | Wei, Xun; Luo, Wang; Zhang, Xixi; Yang, Jie; Gui, Guan; Ohtsuki, Tomoaki Differentiable Architecture Search-Based Automatic Modulation Classification Inproceedings 2021 IEEE Wireless Communications and Networking Conference (WCNC), pp. 1-6, 2021. @inproceedings{9417449, title = {Differentiable Architecture Search-Based Automatic Modulation Classification}, author = {Xun Wei and Wang Luo and Xixi Zhang and Jie Yang and Guan Gui and Tomoaki Ohtsuki}, url = {https://ieeexplore.ieee.org/abstract/document/9417449}, doi = {10.1109/WCNC49053.2021.9417449}, year = {2021}, date = {2021-01-01}, booktitle = {2021 IEEE Wireless Communications and Networking Conference (WCNC)}, pages = {1-6}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1130. | Chen, Haoze; Zhang, Zhijie; Zhao, Chenyang; Liu, Jiaqi; Yin, Wuliang; Li, Yanfeng; Wang, Fengxiang; Li, Chao; Lin, Zhenyu Depth Classification of Defects Based on Neural Architecture Search Journal Article IEEE Access, 9 , pp. 73424-73432, 2021. @article{9424564, title = {Depth Classification of Defects Based on Neural Architecture Search}, author = {Haoze Chen and Zhijie Zhang and Chenyang Zhao and Jiaqi Liu and Wuliang Yin and Yanfeng Li and Fengxiang Wang and Chao Li and Zhenyu Lin}, url = {https://ieeexplore.ieee.org/abstract/document/9424564}, doi = {10.1109/ACCESS.2021.3077961}, year = {2021}, date = {2021-01-01}, journal = {IEEE Access}, volume = {9}, pages = {73424-73432}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1129. | Mo, Hyunho; Custode, Leonardo Lucio; Iacca, Giovanni Evolutionary neural architecture search for remaining useful life prediction Journal Article Applied Soft Computing, 108 , pp. 107474, 2021, ISSN: 1568-4946. @article{MO2021107474, title = {Evolutionary neural architecture search for remaining useful life prediction}, author = {Hyunho Mo and Leonardo Lucio Custode and Giovanni Iacca}, url = {https://www.sciencedirect.com/science/article/pii/S1568494621003975}, doi = {https://doi.org/10.1016/j.asoc.2021.107474}, issn = {1568-4946}, year = {2021}, date = {2021-01-01}, journal = {Applied Soft Computing}, volume = {108}, pages = {107474}, abstract = {With the advent of Industry 4.0, making accurate predictions of the remaining useful life (RUL) of industrial components has become a crucial aspect in predictive maintenance (PdM). To this aim, various Deep Neural Network (DNN) models have been proposed in the recent literature. However, while the architectures of these models have a large impact on their performance, they are usually determined empirically. To exclude the time-consuming process and the unnecessary computational cost of manually engineering these models, we present a Neural Architecture Search (NAS) technique based on an Evolutionary Algorithm (EA) applied to optimize the architecture of a DNN used to predict the RUL. The EA explores the combinatorial parameter space of a multi-head Convolutional Neural Network with Long Short Term Memory (CNN-LSTM) to search for the best architecture. In particular, our method requires minimum computational resources by making use of an early stopping policy and a history of the evaluated architectures. We dub the proposed method ENAS-PdM. To our knowledge, this is the first work where an EA-based NAS is used to optimize a CNN-LSTM architecture in the field of PdM. In our experiments, we use the well-established Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset from NASA. Compared to the current state-of-the-art, our method obtains better results in terms of two different metrics, RMSE and Score, when aggregating across all the C-MAPSS sub-datasets. Without aggregation, we achieve lower RMSE in 3 out of 4 sub-datasets. Our experimental results verify that the proposed method is a reliable tool for obtaining state-of-the-art RUL predictions and as such it can have a strong impact in several industrial applications, especially those with limited available computing power.}, keywords = {}, pubstate = {published}, tppubtype = {article} } With the advent of Industry 4.0, making accurate predictions of the remaining useful life (RUL) of industrial components has become a crucial aspect in predictive maintenance (PdM). To this aim, various Deep Neural Network (DNN) models have been proposed in the recent literature. However, while the architectures of these models have a large impact on their performance, they are usually determined empirically. To exclude the time-consuming process and the unnecessary computational cost of manually engineering these models, we present a Neural Architecture Search (NAS) technique based on an Evolutionary Algorithm (EA) applied to optimize the architecture of a DNN used to predict the RUL. The EA explores the combinatorial parameter space of a multi-head Convolutional Neural Network with Long Short Term Memory (CNN-LSTM) to search for the best architecture. In particular, our method requires minimum computational resources by making use of an early stopping policy and a history of the evaluated architectures. We dub the proposed method ENAS-PdM. To our knowledge, this is the first work where an EA-based NAS is used to optimize a CNN-LSTM architecture in the field of PdM. In our experiments, we use the well-established Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset from NASA. Compared to the current state-of-the-art, our method obtains better results in terms of two different metrics, RMSE and Score, when aggregating across all the C-MAPSS sub-datasets. Without aggregation, we achieve lower RMSE in 3 out of 4 sub-datasets. Our experimental results verify that the proposed method is a reliable tool for obtaining state-of-the-art RUL predictions and as such it can have a strong impact in several industrial applications, especially those with limited available computing power. |
| 1128. | Lv, Jindi; Ye, Qing; Sun, Yanan; Zhao, Juan; Lv, Jiancheng Heart-Darts: Classification of Heartbeats Using Differentiable Architecture Search Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2105-00693, title = {Heart-Darts: Classification of Heartbeats Using Differentiable Architecture Search}, author = {Jindi Lv and Qing Ye and Yanan Sun and Juan Zhao and Jiancheng Lv}, url = {https://arxiv.org/abs/2105.00693}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2105.00693}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1127. | -, Julia Guerrero; Hauns, Sven; Izquierdo, Sergio; Miotto, Guilherme; Schrodi, Simon; Biedenkapp, Andre; Elsken, Thomas; Deng, Difan; Lindauer, Marius; Hutter, Frank Bag of Baselines for Multi-objective Joint Neural Architecture Search and Hyperparameter Optimization Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2105-01015, title = {Bag of Baselines for Multi-objective Joint Neural Architecture Search and Hyperparameter Optimization}, author = {Julia Guerrero - and Sven Hauns and Sergio Izquierdo and Guilherme Miotto and Simon Schrodi and Andre Biedenkapp and Thomas Elsken and Difan Deng and Marius Lindauer and Frank Hutter}, url = {https://arxiv.org/abs/2105.01015}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2105.01015}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1126. | Valsesia, Diego ; Fracastoro, Giulia ; Magli, Enrico Don’t Stack Layers In Graph Neural Networks, Wire Them Randomly Inproceedings ICLR 2021 Workshop on Geometrical and Topological Representation Learning, 2021. @inproceedings{Valsesia2021, title = {Don’t Stack Layers In Graph Neural Networks, Wire Them Randomly}, author = {Valsesia, Diego and Fracastoro, Giulia and Magli, Enrico}, url = {https://openreview.net/pdf/224ad9f05a62ed1b5c7a1a23386bdcd29f717b8a.pdf}, year = {2021}, date = {2021-01-01}, booktitle = {ICLR 2021 Workshop on Geometrical and Topological Representation Learning}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1125. | Bruggemann, David; Kanakis, Menelaos; Obukhov, Anton; Georgoulis, Stamatios; Gool, Luc Van Exploring Relational Context for Multi-Task Dense Prediction Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-13874, title = {Exploring Relational Context for Multi-Task Dense Prediction}, author = {David Bruggemann and Menelaos Kanakis and Anton Obukhov and Stamatios Georgoulis and Luc Van Gool}, url = {https://arxiv.org/abs/2104.13874}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.13874}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1124. | Zhang, Lingling; Wang, Shaowei; Chang, Xiaojun; Liu, Jun; Ge, Zongyuan; Zheng, Qinghua Auto-FSL: Searching the Attribute Consistent Network for Few-Shot Learning Journal Article IEEE Transactions on Circuits and Systems for Video Technology, pp. 1-1, 2021. @article{9419063, title = {Auto-FSL: Searching the Attribute Consistent Network for Few-Shot Learning}, author = {Lingling Zhang and Shaowei Wang and Xiaojun Chang and Jun Liu and Zongyuan Ge and Qinghua Zheng}, url = {https://ieeexplore.ieee.org/abstract/document/9419063}, doi = {10.1109/TCSVT.2021.3076523}, year = {2021}, date = {2021-01-01}, journal = {IEEE Transactions on Circuits and Systems for Video Technology}, pages = {1-1}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1123. | Zhou, Zheng; Li, Tianfu; Zhang, Zilong; Zhao, Zhibin; Sun, Chuang; Yan, Ruqiang; Chen, Xuefeng Bayesian Differentiable Architecture Search for Efficient Domain Matching Fault Diagnosis Journal Article IEEE Transactions on Instrumentation and Measurement, 70 , pp. 1-11, 2021. @article{9419078, title = {Bayesian Differentiable Architecture Search for Efficient Domain Matching Fault Diagnosis}, author = {Zheng Zhou and Tianfu Li and Zilong Zhang and Zhibin Zhao and Chuang Sun and Ruqiang Yan and Xuefeng Chen}, url = {https://ieeexplore.ieee.org/abstract/document/9419078}, doi = {10.1109/TIM.2021.3076575}, year = {2021}, date = {2021-01-01}, journal = {IEEE Transactions on Instrumentation and Measurement}, volume = {70}, pages = {1-11}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1122. | Yang, Hong; Zhang, Ya-sheng; Yin, Can-bin; Ding, Wen-zhe Defence Technology, 2021, ISSN: 2214-9147. @article{YANG2021, title = {Ultra-lightweight CNN design based on neural architecture search and knowledge distillation: A novel method to build the automatic recognition model of space target ISAR images}, author = {Hong Yang and Ya-sheng Zhang and Can-bin Yin and Wen-zhe Ding}, url = {https://www.sciencedirect.com/science/article/pii/S2214914721000763}, doi = {https://doi.org/10.1016/j.dt.2021.04.014}, issn = {2214-9147}, year = {2021}, date = {2021-01-01}, journal = {Defence Technology}, abstract = {In this paper, a novel method of ultra-lightweight convolution neural network (CNN) design based on neural architecture search (NAS) and knowledge distillation (KD) is proposed. It can realize the automatic construction of the space target inverse synthetic aperture radar (ISAR) image recognition model with ultra-lightweight and high accuracy. This method introduces the NAS method into the radar image recognition for the first time, which solves the time-consuming and labor-consuming problems in the artificial design of the space target ISAR image automatic recognition model (STIIARM). On this basis, the NAS model's knowledge is transferred to the student model with lower computational complexity by the flow of the solution procedure (FSP) distillation method. Thus, the decline of recognition accuracy caused by the direct compression of model structural parameters can be effectively avoided, and the ultra-lightweight STIIARM can be obtained. In the method, the Inverted Linear Bottleneck (ILB) and Inverted Residual Block (IRB) are firstly taken as each block's basic structure in CNN. And the expansion ratio, output filter size, number of IRBs, and convolution kernel size are set as the search parameters to construct a hierarchical decomposition search space. Then, the recognition accuracy and computational complexity are taken as the objective function and constraint conditions, respectively, and the global optimization model of the CNN architecture search is established. Next, the simulated annealing (SA) algorithm is used as the search strategy to search out the lightweight and high accuracy STIIARM directly. After that, based on the three principles of similar block structure, the same corresponding channel number, and the minimum computational complexity, the more lightweight student model is designed, and the FSP matrix pairing between the NAS model and student model is completed. Finally, by minimizing the loss between the FSP matrix pairs of the NAS model and student model, the student model's weight adjustment is completed. Thus the ultra-lightweight and high accuracy STIIARM is obtained. The proposed method's effectiveness is verified by the simulation experiments on the ISAR image dataset of five types of space targets.}, keywords = {}, pubstate = {published}, tppubtype = {article} } In this paper, a novel method of ultra-lightweight convolution neural network (CNN) design based on neural architecture search (NAS) and knowledge distillation (KD) is proposed. It can realize the automatic construction of the space target inverse synthetic aperture radar (ISAR) image recognition model with ultra-lightweight and high accuracy. This method introduces the NAS method into the radar image recognition for the first time, which solves the time-consuming and labor-consuming problems in the artificial design of the space target ISAR image automatic recognition model (STIIARM). On this basis, the NAS model's knowledge is transferred to the student model with lower computational complexity by the flow of the solution procedure (FSP) distillation method. Thus, the decline of recognition accuracy caused by the direct compression of model structural parameters can be effectively avoided, and the ultra-lightweight STIIARM can be obtained. In the method, the Inverted Linear Bottleneck (ILB) and Inverted Residual Block (IRB) are firstly taken as each block's basic structure in CNN. And the expansion ratio, output filter size, number of IRBs, and convolution kernel size are set as the search parameters to construct a hierarchical decomposition search space. Then, the recognition accuracy and computational complexity are taken as the objective function and constraint conditions, respectively, and the global optimization model of the CNN architecture search is established. Next, the simulated annealing (SA) algorithm is used as the search strategy to search out the lightweight and high accuracy STIIARM directly. After that, based on the three principles of similar block structure, the same corresponding channel number, and the minimum computational complexity, the more lightweight student model is designed, and the FSP matrix pairing between the NAS model and student model is completed. Finally, by minimizing the loss between the FSP matrix pairs of the NAS model and student model, the student model's weight adjustment is completed. Thus the ultra-lightweight and high accuracy STIIARM is obtained. The proposed method's effectiveness is verified by the simulation experiments on the ISAR image dataset of five types of space targets. |
| 1121. | Yan, Bin; Peng, Houwen; Wu, Kan; Wang, Dong; Fu, Jianlong; Lu, Huchuan LightTrack: Finding Lightweight Neural Networks for Object Tracking via One-Shot Architecture Search Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-14545b, title = {LightTrack: Finding Lightweight Neural Networks for Object Tracking via One-Shot Architecture Search}, author = {Bin Yan and Houwen Peng and Kan Wu and Dong Wang and Jianlong Fu and Huchuan Lu}, url = {https://arxiv.org/abs/2104.14545}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.14545}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1120. | Oymak, Samet; Li, Mingchen; Soltanolkotabi, Mahdi Generalization Guarantees for Neural Architecture Search with Train-Validation Split Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-14132, title = {Generalization Guarantees for Neural Architecture Search with Train-Validation Split}, author = {Samet Oymak and Mingchen Li and Mahdi Soltanolkotabi}, url = {https://arxiv.org/abs/2104.14132}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.14132}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1119. | Byun, Younghoon; Lee, Youngjoo Rapid Design Space Exploration of Near-Optimal Memory-Reduced DCNN Architecture using Multiple Model Compression Techniques Inproceedings 2021 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1-5, 2021. @inproceedings{9401489, title = {Rapid Design Space Exploration of Near-Optimal Memory-Reduced DCNN Architecture using Multiple Model Compression Techniques}, author = {Younghoon Byun and Youngjoo Lee}, url = {https://ieeexplore.ieee.org/abstract/document/9401489}, doi = {10.1109/ISCAS51556.2021.9401489}, year = {2021}, date = {2021-01-01}, booktitle = {2021 IEEE International Symposium on Circuits and Systems (ISCAS)}, pages = {1-5}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1118. | Risso, Matteo; Burrello, Alessio; Pagliari, Daniele Jahier; Benatti, Simone; Macii, Enrico; Benini, Luca; Poncino, Massimo Robust and Energy-efficient PPG-based Heart-Rate Monitoring Inproceedings 2021 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1-5, 2021. @inproceedings{9401282, title = {Robust and Energy-efficient PPG-based Heart-Rate Monitoring}, author = {Matteo Risso and Alessio Burrello and Daniele Jahier Pagliari and Simone Benatti and Enrico Macii and Luca Benini and Massimo Poncino}, url = {https://ieeexplore.ieee.org/abstract/document/9401282}, doi = {10.1109/ISCAS51556.2021.9401282}, year = {2021}, date = {2021-01-01}, booktitle = {2021 IEEE International Symposium on Circuits and Systems (ISCAS)}, pages = {1-5}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1117. | Chen, Oscal Tzyh-Chiang; Zhang, Yu Cheng; Chang, Yu-Xuan; Chang, Yu-Lung Iterative Multiple-Path One-Shot NAS for the Optimized Performance Inproceedings 2021 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1-5, 2021. @inproceedings{9401130, title = {Iterative Multiple-Path One-Shot NAS for the Optimized Performance}, author = {Oscal Tzyh-Chiang Chen and Yu Cheng Zhang and Yu-Xuan Chang and Yu-Lung Chang}, url = {https://ieeexplore.ieee.org/abstract/document/9401130}, doi = {10.1109/ISCAS51556.2021.9401130}, year = {2021}, date = {2021-01-01}, booktitle = {2021 IEEE International Symposium on Circuits and Systems (ISCAS)}, pages = {1-5}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1116. | Laube, Kevin Alexander; Zell, Andreas Inter-choice dependent super-network weights Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-11522, title = {Inter-choice dependent super-network weights}, author = {Kevin Alexander Laube and Andreas Zell}, url = {https://arxiv.org/abs/2104.11522}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.11522}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1115. | Lei, Zhou; Yang, Kangkang; Jiang, Kai; Chen, Shengbo KDAS-ReID: Architecture Search for Person Re-Identification via Distilled Knowledge with Dynamic Temperature Journal Article Algorithms, 14 (5), 2021, ISSN: 1999-4893. @article{a14050137, title = {KDAS-ReID: Architecture Search for Person Re-Identification via Distilled Knowledge with Dynamic Temperature}, author = {Zhou Lei and Kangkang Yang and Kai Jiang and Shengbo Chen}, url = {https://www.mdpi.com/1999-4893/14/5/137}, doi = {10.3390/a14050137}, issn = {1999-4893}, year = {2021}, date = {2021-01-01}, journal = {Algorithms}, volume = {14}, number = {5}, abstract = {Person re-Identification(Re-ID) based on deep convolutional neural networks (CNNs) achieves remarkable success with its fast speed. However, prevailing Re-ID models are usually built upon backbones that manually design for classification. In order to automatically design an effective Re-ID architecture, we propose a pedestrian re-identification algorithm based on knowledge distillation, called KDAS-ReID. When the knowledge of the teacher model is transferred to the student model, the importance of knowledge in the teacher model will gradually decrease with the improvement of the performance of the student model. Therefore, instead of applying the distillation loss function directly, we consider using dynamic temperatures during the search stage and training stage. Specifically, we start searching and training at a high temperature and gradually reduce the temperature to 1 so that the student model can better learn from the teacher model through soft targets. Extensive experiments demonstrate that KDAS-ReID performs not only better than other state-of-the-art Re-ID models on three benchmarks, but also better than the teacher model based on the ResNet-50 backbone.}, keywords = {}, pubstate = {published}, tppubtype = {article} } Person re-Identification(Re-ID) based on deep convolutional neural networks (CNNs) achieves remarkable success with its fast speed. However, prevailing Re-ID models are usually built upon backbones that manually design for classification. In order to automatically design an effective Re-ID architecture, we propose a pedestrian re-identification algorithm based on knowledge distillation, called KDAS-ReID. When the knowledge of the teacher model is transferred to the student model, the importance of knowledge in the teacher model will gradually decrease with the improvement of the performance of the student model. Therefore, instead of applying the distillation loss function directly, we consider using dynamic temperatures during the search stage and training stage. Specifically, we start searching and training at a high temperature and gradually reduce the temperature to 1 so that the student model can better learn from the teacher model through soft targets. Extensive experiments demonstrate that KDAS-ReID performs not only better than other state-of-the-art Re-ID models on three benchmarks, but also better than the teacher model based on the ResNet-50 backbone. |
| 1114. | Amer, Mohammed; á, Tom; Liao, Iman Yi Balancing Accuracy and Latency in Multipath Neural Networks Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-12040, title = {Balancing Accuracy and Latency in Multipath Neural Networks}, author = {Mohammed Amer and Tom á and Iman Yi Liao}, url = {https://arxiv.org/abs/2104.12040}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.12040}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1113. | Yang, Zhao; Sun, Qingshuang Efficient Resource-aware Neural Architecture Search with Dynamic Adaptive Network Sampling Inproceedings 2021 IEEE International Symposium on Circuits and Systems (ISCAS), pp. 1-5, 2021. @inproceedings{9401187, title = {Efficient Resource-aware Neural Architecture Search with Dynamic Adaptive Network Sampling}, author = {Zhao Yang and Qingshuang Sun}, url = {https://ieeexplore.ieee.org/abstract/document/9401187}, doi = {10.1109/ISCAS51556.2021.9401187}, year = {2021}, date = {2021-01-01}, booktitle = {2021 IEEE International Symposium on Circuits and Systems (ISCAS)}, pages = {1-5}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1112. | Pasini, Massimiliano Lupo; Yin, Junqi; Li, Ying Wai; Eisenbach, Markus A scalable algorithm for the optimization of neural network architectures Journal Article Parallel Computing, 104-105 , pp. 102788, 2021, ISSN: 0167-8191. @article{LUPOPASINI2021102788, title = {A scalable algorithm for the optimization of neural network architectures}, author = {Massimiliano Lupo Pasini and Junqi Yin and Ying Wai Li and Markus Eisenbach}, url = {https://www.sciencedirect.com/science/article/pii/S0167819121000430}, doi = {https://doi.org/10.1016/j.parco.2021.102788}, issn = {0167-8191}, year = {2021}, date = {2021-01-01}, journal = {Parallel Computing}, volume = {104-105}, pages = {102788}, abstract = {We propose a new scalable method to optimize the architecture of an artificial neural network. The proposed algorithm, called Greedy Search for Neural Network Architecture, aims to determine a neural network with minimal number of layers that is at least as performant as neural networks of the same structure identified by other hyperparameter search algorithms in terms of accuracy and computational cost. Numerical results performed on benchmark datasets show that, for these datasets, our method outperforms state-of-the-art hyperparameter optimization algorithms in terms of attainable predictive performance by the selected neural network architecture, and time-to-solution for the hyperparameter optimization to complete.}, keywords = {}, pubstate = {published}, tppubtype = {article} } We propose a new scalable method to optimize the architecture of an artificial neural network. The proposed algorithm, called Greedy Search for Neural Network Architecture, aims to determine a neural network with minimal number of layers that is at least as performant as neural networks of the same structure identified by other hyperparameter search algorithms in terms of accuracy and computational cost. Numerical results performed on benchmark datasets show that, for these datasets, our method outperforms state-of-the-art hyperparameter optimization algorithms in terms of attainable predictive performance by the selected neural network architecture, and time-to-solution for the hyperparameter optimization to complete. |
| 1111. | Clemens, John MLDS: A Dataset for Weight-Space Analysis of Neural Networks Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-10555, title = {MLDS: A Dataset for Weight-Space Analysis of Neural Networks}, author = {John Clemens}, url = {https://arxiv.org/abs/2104.10555}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.10555}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1110. | Di, Shimin; Yao, Quanming; Zhang, Yongqi; Chen, Lei Efficient Relation-aware Scoring Function Search for Knowledge Graph Embedding Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-10880, title = {Efficient Relation-aware Scoring Function Search for Knowledge Graph Embedding}, author = {Shimin Di and Quanming Yao and Yongqi Zhang and Lei Chen}, url = {https://arxiv.org/abs/2104.10880}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.10880}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1109. | Di, Shimin; Yao, Quanming; Chen, Lei Searching to Sparsify Tensor Decomposition for N-ary Relational Data Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-10625, title = {Searching to Sparsify Tensor Decomposition for N-ary Relational Data}, author = {Shimin Di and Quanming Yao and Lei Chen}, url = {https://arxiv.org/abs/2104.10625}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.10625}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1108. | Duggal, Rahul; Zhou, Hao; Yang, Shuo; Xiong, Yuanjun; Xia, Wei; Tu, Zhuowen; Soatto, Stefano Compatibility-aware Heterogeneous Visual Search Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2105-06047, title = {Compatibility-aware Heterogeneous Visual Search}, author = {Rahul Duggal and Hao Zhou and Shuo Yang and Yuanjun Xiong and Wei Xia and Zhuowen Tu and Stefano Soatto}, url = {https://arxiv.org/abs/2105.06047}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2105.06047}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1107. | -, Min; -, Hao; -, Min; -, Yu; -, Hsien; -, Yi; -, Hung; Jou, Kevin Network Space Search for Pareto-Efficient Spaces Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-11014, title = {Network Space Search for Pareto-Efficient Spaces}, author = {Min - and Hao - and Min - and Yu - and Hsien - and Yi - and Hung - and Kevin Jou}, url = {https://arxiv.org/abs/2104.11014}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.11014}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1106. | Bodin, Erik; Tomasi, Federico; Dai, Zhenwen Making Differentiable Architecture Search less local Technical Report , 2021. @techreport{DBLP:journals/corr/abs-2104-10450, title = {Making Differentiable Architecture Search less local}, author = {Erik Bodin and Federico Tomasi and Zhenwen Dai}, url = {https://arxiv.org/abs/2104.10450}, year = {2021}, date = {2021-01-01}, journal = {CoRR}, volume = {abs/2104.10450}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1105. | Song, Seok Bin; Nam, Jung Woo; Kim, Jin Heon NAS-PPG: PPG based heart rate estimation using Neural Architecture Search Journal Article IEEE Sensors Journal, pp. 1-1, 2021. @article{9402889, title = {NAS-PPG: PPG based heart rate estimation using Neural Architecture Search}, author = {Seok Bin Song and Jung Woo Nam and Jin Heon Kim}, url = {https://ieeexplore.ieee.org/abstract/document/9402889}, doi = {10.1109/JSEN.2021.3073047}, year = {2021}, date = {2021-01-01}, journal = {IEEE Sensors Journal}, pages = {1-1}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1104. | Wu, Haiwei; Zhou, Jiantao IID-Net: Image Inpainting Detection Network via Neural Architecture Search and Attention Journal Article IEEE Transactions on Circuits and Systems for Video Technology, pp. 1-1, 2021. @article{9410590, title = {IID-Net: Image Inpainting Detection Network via Neural Architecture Search and Attention}, author = {Haiwei Wu and Jiantao Zhou}, url = {https://ieeexplore.ieee.org/abstract/document/9410590}, doi = {10.1109/TCSVT.2021.3075039}, year = {2021}, date = {2021-01-01}, journal = {IEEE Transactions on Circuits and Systems for Video Technology}, pages = {1-1}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1103. | Verma, Monu; Reddy, Satish Kumar M; Meedimale, Yashwanth Reddy; Mandal, Murari; Vipparthi, Santosh Kumar AutoMER: Spatiotemporal Neural Architecture Search for Microexpression Recognition Journal Article IEEE Transactions on Neural Networks and Learning Systems, pp. 1-13, 2021. @article{9411707, title = {AutoMER: Spatiotemporal Neural Architecture Search for Microexpression Recognition}, author = {Monu Verma and Satish Kumar M Reddy and Yashwanth Reddy Meedimale and Murari Mandal and Santosh Kumar Vipparthi}, url = {https://ieeexplore.ieee.org/abstract/document/9411707/}, doi = {10.1109/TNNLS.2021.3072290}, year = {2021}, date = {2021-01-01}, journal = {IEEE Transactions on Neural Networks and Learning Systems}, pages = {1-13}, keywords = {}, pubstate = {published}, tppubtype = {article} } |
| 1102. | Chen, Minghao; Peng, Houwen; Fu, Jianlong; Ling, Haibin One-Shot Neural Ensemble Architecture Search by Diversity-Guided Search Space Shrinking Inproceedings CVPR2021, 2021. @inproceedings{DBLP:journals/corr/abs-2104-00597b, title = {One-Shot Neural Ensemble Architecture Search by Diversity-Guided Search Space Shrinking}, author = {Minghao Chen and Houwen Peng and Jianlong Fu and Haibin Ling}, url = {https://arxiv.org/abs/2104.00597}, year = {2021}, date = {2021-01-01}, booktitle = {CVPR2021}, journal = {CoRR}, volume = {abs/2104.00597}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1101. | Yan, Bin; Peng, Houwen; Wu, Kan; Wang, Dong; Fu, Jianlong; Lu, Huchuan LightTrack: Finding Lightweight Neural Networks for Object Tracking via One-Shot Architecture Search Inproceedings CVPR2021, 2021. @inproceedings{DBLP:journals/corr/abs-2104-14545, title = {LightTrack: Finding Lightweight Neural Networks for Object Tracking via One-Shot Architecture Search}, author = {Bin Yan and Houwen Peng and Kan Wu and Dong Wang and Jianlong Fu and Huchuan Lu}, url = {https://arxiv.org/abs/2104.14545}, year = {2021}, date = {2021-01-01}, booktitle = {CVPR2021}, journal = {CoRR}, volume = {abs/2104.14545}, keywords = {}, pubstate = {published}, tppubtype = {inproceedings} } |
| 1100. | Wang, Jiahao; Shu, Han; Xia, Weihao; Yang, Yujiu; Wang, Yunhe Coarse-to-Fine Searching for Efficient Generative Adversarial Networks Technical Report 2021. @techreport{wang2021coarsetofine, title = {Coarse-to-Fine Searching for Efficient Generative Adversarial Networks}, author = {Jiahao Wang and Han Shu and Weihao Xia and Yujiu Yang and Yunhe Wang}, url = {https://arxiv.org/abs/2104.09223}, year = {2021}, date = {2021-01-01}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1099. | Ahn, Joon Young; Cho, Nam Ik Neural Architecture Search for Image Super-Resolution Using Densely Constructed Search Space: DeCoNAS Technical Report 2021. @techreport{ahn2021neural, title = {Neural Architecture Search for Image Super-Resolution Using Densely Constructed Search Space: DeCoNAS}, author = {Joon Young Ahn and Nam Ik Cho}, url = {https://arxiv.org/abs/2104.09048}, year = {2021}, date = {2021-01-01}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |
| 1098. | Yin, Yihang; Huang, Siyu; Zhang, Xiang; Dou, Dejing BM-NAS: Bilevel Multimodal Neural Architecture Search Technical Report 2021. @techreport{yin2021bmnas, title = {BM-NAS: Bilevel Multimodal Neural Architecture Search}, author = {Yihang Yin and Siyu Huang and Xiang Zhang and Dejing Dou}, url = {https://arxiv.org/abs/2104.09379}, year = {2021}, date = {2021-01-01}, keywords = {}, pubstate = {published}, tppubtype = {techreport} } |