Literature on Neural Architecture Search

@article{Stanley2009_mid,

title = {A Hypercube-Based Encoding for Evolving Large-Scale Neural Networks},

author = {Kenneth O Stanley and David B D'Ambrosio and Jason Gauci},

url = {https://ieeexplore.ieee.org/document/6792316/},

doi = {10.1162/ARTL.2009.15.2.15202},

year  = {2009},

date = {2009-01-01},

volume = {15},

number = {2},

pages = {185-212},

key = {journals/alife/StanleyDG09},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Floreano2008_wpi,

title = {Neuroevolution - from architectures to learning},

author = {Dario Floreano and Peter Dürr and Claudio Mattiussi},

url = {https://link.springer.com/article/10.1007/s12065-007-0002-4},

doi = {10.1007/S12065-007-0002-4},

year  = {2008},

date = {2008-01-01},

volume = {1},

number = {1},

pages = {47-62},

key = {journals/evi/FloreanoDM08},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Roy2004_app,

title = {Designing Polymer Blends Using Neural Networks, Genetic Algorithms, and Markov Chains},

author = {N K Roy and Walter D Potter and D P Landau},

url = {https://dl.acm.org/citation.cfm?id=94034},

doi = {10.1023/B:APIN.0000021414.50728.34},

year  = {2004},

date = {2004-01-01},

volume = {20},

number = {3},

pages = {215-229},

key = {journals/apin/RoyPL04},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Angeline1994_qts,

title = {An evolutionary algorithm that constructs recurrent neural networks},

author = {Peter J Angeline and Gregory M Saunders and Jordan B Pollack},

url = {https://ieeexplore.ieee.org/document/265960/},

doi = {10.1109/72.265960},

year  = {1994},

date = {1994-01-01},

volume = {5},

number = {1},

pages = {54-65},

key = {journals/tnn/AngelineSP94},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Kitano1990_uli,

title = {Designing Neural Networks Using Genetic Algorithms with Graph Generation System},

author = {Hiroaki Kitano},

url = {http://www.complex-systems.com/abstracts/v04_i04_a06/},

year  = {1990},

date = {1990-01-01},

volume = {4},

number = {4},

key = {journals/compsys/Kitano90},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{doi:10.1142/S1793351X22500027,

title = {Evolving Deep Neural Networks with Cultural Algorithms for Real-Time Industrial Applications},

author = {Faisal Waris and Robert G. Reynolds and Joonho Lee},

url = {https://doi.org/10.1142/S1793351X22500027},

doi = {10.1142/S1793351X22500027},

year  = {0000},

date = {0000-01-01},

urldate = {0000-01-01},

journal = {International Journal of Semantic Computing},

volume = {0},

number = {0},

pages = {1-32},

abstract = {The goal of this paper is to investigate the applicability of evolutionary algorithms to the design of real-time industrial controllers. Present-day “deep learning” (DL) is firmly established as a useful tool for addressing many practical problems. This has spurred the development of neural architecture search (NAS) methods in order to automate the model search activity. CATNeuro is a NAS algorithm based on the graph evolution concept devised by Neuroevolution of Augmenting Topologies (NEAT) but propelled by cultural algorithm (CA) as the evolutionary driver. The CA is a network-based, stochastic optimization framework inspired by problem solving in human cultures. Knowledge distribution (KD) across the network of graph models is a key to problem solving success in CAT systems. Two alternative mechanisms for KD across the network are employed. One supports cooperation (CATNeuro) in the network and the other competition (WM). To test the viability of each configuration prior to use in the industrial setting, they were applied to the design of a real-time controller for a two-dimensional fighting game. While both were able to beat the AI program that came with the fighting game, the cooperative method performed statistically better. As a result, it was used to track the motion of a trailer (in lateral and vertical directions) using a camera mounted on the tractor vehicle towing the trailer. In this second real-time application (trailer motion), the CATNeuro configuration was compared to the original NEAT (elitist) method of evolution. CATNeuro is found to perform statistically better than NEAT in many aspects of the design including model training loss, model parameter size, and overall model structure consistency. In both scenarios, the performance improvements were attributed to the increased model diversity due to the interaction of CA knowledge sources both cooperatively and competitively.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{https://doi.org/10.1111/exsy.12690,

title = {Hybrid evolutionary network architecture search (HyENAS) for convolution class of deep neural networks with applications},

author = {Soniya and Lotika Singh and Sandeep Paul},

url = {https://onlinelibrary.wiley.com/doi/abs/10.1111/exsy.12690},

doi = {https://doi.org/10.1111/exsy.12690},

journal = {Expert Systems},

volume = {n/a},

number = {n/a},

pages = {e12690},

abstract = {Abstract Convolutional Neural Networks (CNNs) and its variants are increasingly used across wide domain of applications achieving high performance measures. For high performance, application specific CNN architecture is required, hence the need for network architecture search (NAS) becomes essential. This paper proposes a hybrid evolutionary approach for network architecture search (HyENAS), and targets convolution class of neural networks. One of the significant contribution of this technique is to completely evolve the high performance network by simultaneously finding network structures and their corresponding parameters. An elegant string representation has been proposed which efficiently represents the network. The concept of sparse block evolving requisite layer wise features for dense network is deployed. This permits the network to dynamically evolve for a specific application. In comparison to the other state-of-art methods, the high performance of the proposed HyENAS approach is demonstrated across various benchmark data sets belonging to the domain of malariology, oncology, neurology, ophthalmology, and genomics. Further, to deploy the proposed model on lower hardware specification devices, another salient feature of the HyENAS technique is to seamlessly sift out the simpler network architecture with comparable accuracy.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{https://doi.org/10.1049/cvi2.12036,

title = {Control the number of skip-connects to improve robustness of the NAS algorithm},

author = {Bao Feng Zhang and Guo Qiang Zhou},

url = {https://ietresearch.onlinelibrary.wiley.com/doi/abs/10.1049/cvi2.12036},

doi = {https://doi.org/10.1049/cvi2.12036},

journal = {IET Computer Vision},

volume = {n/a},

number = {n/a},

abstract = {Abstract Recently, the gradient-based neural architecture search has made remarkable progress with the characteristics of high efficiency and fast convergence. However, two common problems in the gradient-based NAS algorithms are found. First, with the increase in the raining time, the NAS algorithm tends to skip-connect operation, leading to performance degradation and instability results. Second, another problem is no reasonable allocation of computing resources on valuable candidate network models. The above two points lead to the difficulty in searching the optimal sub-network and poor stability. To address them, the trick of pre-training the super-net is applied, so that each operation has an equal opportunity to develop its strength, which provides a fair competition condition for the convergence of the architecture parameters. In addition, a skip-controller is proposed to ensure each sampled sub-network with an appropriate number of skip-connects. The experiments were performed on three mainstream datasets CIFAR-10, CIFAR-100 and ImageNet, in which the improved method achieves comparable results with higher accuracy and stronger robustness.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{https://doi.org/10.1002/int.22471,

title = {License plate recognition using neural architecture search for edge devices},

author = {Jithmi Shashirangana and Heshan Padmasiri and Dulani Meedeniya and Charith Perera and Soumya R Nayak and Janmenjoy Nayak and Shanmuganthan Vimal and Seifidine Kadry},

url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/int.22471},

doi = {https://doi.org/10.1002/int.22471},

journal = {International Journal of Intelligent Systems},

volume = {n/a},

number = {n/a},

abstract = {Abstract The mutually beneficial blend of artificial intelligence with internet of things has been enabling many industries to develop smart information processing solutions. The implementation of technology enhanced industrial intelligence systems is challenging with the environmental conditions, resource constraints and safety concerns. With the era of smart homes and cities, domains like automated license plate recognition (ALPR) are exploring automate tasks such as traffic management and fraud detection. This paper proposes an optimized decision support solution for ALPR that works purely on edge devices at night-time. Although ALPR is a frequently addressed research problem in the domain of intelligent systems, still they are generally computationally intensive and unable to run on edge devices with limited resources. Therefore, as a novel approach, we consider the complex aspects related to deploying lightweight yet efficient and fast ALPR models on embedded devices. The usability of the proposed models is assessed in real-world with a proof-of-concept hardware design and achieved competitive results to the state-of-the-art ALPR solutions that run on server-grade hardware with intensive resources.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{doi:10.1142/S0129065721500295,

title = {One-Shot Neural Architecture Search by Dynamically Pruning Supernet in Hierarchical Order},

author = {Jianwei Zhang and Dong Li and Lituan Wang and Lei Zhang},

url = {https://doi.org/10.1142/S0129065721500295},

doi = {10.1142/S0129065721500295},

journal = {International Journal of Neural Systems},

abstract = {Neural Architecture Search (NAS), which aims at automatically designing neural architectures, recently draw a growing research interest. Different from conventional NAS methods, in which a large number of neural architectures need to be trained for evaluation, the one-shot NAS methods only have to train one supernet which synthesizes all the possible candidate architectures. As a result, the search efficiency could be significantly improved by sharing the supernet’s weights during the candidate architectures’ evaluation. This strategy could greatly speed up the search process but suffer a challenge that the evaluation based on sharing weights is not predictive enough. Recently, pruning the supernet during the search has been proven to be an efficient way to alleviate this problem. However, the pruning direction in complex-structured search space remains unexplored. In this paper, we revisited the role of path dropout strategy, which drops the neural operations instead of the neurons, in supernet training, and several interesting characters of the supernet trained with dropout are found. Based on the observations, a Hierarchically-Ordered Pruning Neural Architecture Search (HOPNAS) algorithm is proposed by dynamically pruning the supernet with a proper pruning direction. Experimental results indicate that our method is competitive with state-of-the-art approaches on CIFAR10 and ImageNet.},

note = {PMID: 34128778},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Sapra2021,

title = {Designing convolutional neural networks with constrained evolutionary piecemeal training},

author = {Dolly Sapra and Andy D. Pimentel },

url = {https://link.springer.com/article/10.1007/s10489-021-02679-7},

journal = {Applied Intelligence },

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Artin21,

title = {Presentation of a Novel Method for Prediction of Traffic with Climate Condition Based on Ensemble Learning of Neural Architecture Search (NAS) and Linear Regression},

author = {Javad Artin and Amin Valizadeh and Mohsen Ahmadi and Sathish A. P. Kumar and Abbas Sharifi },

url = {https://doi.org/10.1155/2021/8500572},

journal = {Complexity},

abstract = {Traffic prediction is critical to expanding a smart city and country because it improves urban planning and traffic management. This prediction is very challenging due to the multifactorial and random nature of traffic. This study presented a method based on ensemble learning to predict urban traffic congestion based on weather criteria. We used the NAS algorithm, which in the output based on heuristic methods creates an optimal model concerning input data. We had 400 data, which included the characteristics of the day’s weather, including six features: absolute humidity, dew point, visibility, wind speed, cloud height, and temperature, which in the final column is the urban traffic congestion target. We have analyzed linear regression with the results obtained in the project; this method was more efficient than other regression models. This method had an error of 0.00002 in terms of MSE criteria and SVR, random forest, and MLP methods, which have error values of 0.01033, 0.00003, and 0.0011, respectively. According to the MAE criterion, this method has a value of 0.0039. The other methods have obtained values of 0.0850, 0.0045, and 0.027, respectively, which show that our proposed model has a minor error than other methods and has been able to outpace the other models.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Gong2021,

title = {Analyze COVID-19 CT images based on evolutionary algorithm with dynamic searching space},

author = {Yunhong Gong and Yanan Sun and Dezhong Peng and Peng Chen and Zhongtai Yan and Ke Yang },

url = {https://link.springer.com/article/10.1007/s40747-021-00513-8},

journal = {Complex & Intelligent Systems },

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{Ma2022,

title = {LPI Radar Waveform Recognition Based on Neural Architecture Search},

author = {Zhiyuan Ma and Wenting Yu and Peng Zhang and Zhi Huang and Anni Lin and Yan Xia},

url = {https://doi.org/10.1155/2022/4628481},

journal = {Computational Intelligence and Neuroscience},

volume = {2022},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{dhanarajaleveraging,

title = {Leveraging Tensor Methods in Neural Architecture Search for the automatic development of lightweight Convolutional Neural Networks},

author = {Mayur Dhanaraja and Huyen Dob and Dinesh Nairb and Cong Xub},

url = {https://assets.amazon.science/34/29/484392a6450b8af9b7646fe6db60/leveraging-tensor-methods-in-neural-architecture-search-for-the-automatic-development-of-lightweight-convolutional-neural-networks.pdf},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{https://doi.org/10.1002/mp.15694,

title = {Improvement of IMRT QA prediction using imaging-based neural architecture search},

author = {Yao Hao and Xizhe Zhang and Jie Wang and Tianyu Zhao and Baozhou Sun},

url = {https://aapm.onlinelibrary.wiley.com/doi/abs/10.1002/mp.15694},

doi = {https://doi.org/10.1002/mp.15694},

journal = {Medical Physics},

volume = {n/a},

number = {n/a},

abstract = {Abstract Purpose Machine learning (ML) has been used to predict the gamma passing rate (GPR) of intensity-modulated radiation therapy (IMRT) QA results. In this work, we applied a novel neural architecture search to automatically tune and search for the best deep neural networks instead of using hand-designed deep learning architectures. Method and materials One hundred and eighty-two IMRT plans were created and delivered with portal dosimetry. A total of 1497 fields for multiple treatment sites were delivered and measured by portal imagers. Gamma criteria of 2%/2 mm with a 5% threshold were used. Fluence maps calculated for each plan were used as inputs to a convolution neural network (CNN). Auto-Keras was implemented to search for the best CNN architecture for fluence image regression. The network morphism was adopted in the searching process, in which the base models were ResNet and DenseNet. The performance of this CNN approach was compared with tree-based ML models previously developed for this application, using the same dataset. Results The deep-learning-based approach had 98.3% of predictions within 3% of the measured 2%/2-mm GPRs with a maximum error of 3.1% and a mean absolute error of less than 1%. Our results show that this novel architecture search approach achieves comparable performance to the machine-learning-based approaches with handcrafted features. Conclusions We implemented a novel CNN model using imaging-based neural architecture for IMRT QA prediction. The imaging-based deep-learning method does not require a manual extraction of relevant features and is able to automatically select the best network architecture.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{https://doi.org/10.1049/rsn2.12273,

title = {Adaptive feature extraction and fine-grained modulation recognition of multi-function radar under small sample conditions},

author = {Qihang Zhai and Yan Li and Zilin Zhang and Yunjie Li and Shafei Wang},

url = {https://ietresearch.onlinelibrary.wiley.com/doi/abs/10.1049/rsn2.12273},

doi = {https://doi.org/10.1049/rsn2.12273},

journal = {IET Radar, Sonar & Navigation},

volume = {n/a},

number = {n/a},

abstract = {Abstract Multi-function radars (MFRs) are sophisticated sensors with fine-grained modes, which modify their modulation types and parameters range generating various signals to fulfil different tasks, such as surveillance and tracking. In electromagnetic reconnaissance, recognition of MFR fine-grained modes can provide a basis for analysing strategies and reaction. With the limit of real applications, it is hard to obtain a large number of labelled samples for existing methods to learn the difference between categories. Therefore, it is essential to develop new methods to extract general knowledge of MFRs and identify modes with only a few samples. This paper proposes a few-shot learning (FSL) framework based on efficient neural architecture search (ENAS) with high robustness and portability, which designs a suitable network structure automated and quickly adapts to new environments. The experimental results show that the proposed method can still achieve excellent fine-grained modulation recognition performance (92.6%) under the condition of -6 dB signal-to-noise ratio (SNR), even when each class only provides one fixed-duration signal sample. The robustness is also verified under different conditions.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{https://doi.org/10.1049/cit2.12106,

title = {Scale adaptive fitness evaluation-based particle swarm optimisation for hyperparameter and architecture optimisation in neural networks and deep learning},

author = {Ye-Qun Wang and Jian-Yu Li and Chun-Hua Chen and Jun Zhang and Zhi-Hui Zhan},

url = {https://ietresearch.onlinelibrary.wiley.com/doi/abs/10.1049/cit2.12106},

doi = {https://doi.org/10.1049/cit2.12106},

journal = {CAAI Transactions on Intelligence Technology},

volume = {n/a},

number = {n/a},

abstract = {Abstract Research into automatically searching for an optimal neural network (NN) by optimisation algorithms is a significant research topic in deep learning and artificial intelligence. However, this is still challenging due to two issues: Both the hyperparameter and architecture should be optimised and the optimisation process is computationally expensive. To tackle these two issues, this paper focusses on solving the hyperparameter and architecture optimization problem for the NN and proposes a novel light-weight scale-adaptive fitness evaluation-based particle swarm optimisation (SAFE-PSO) approach. Firstly, the SAFE-PSO algorithm considers the hyperparameters and architectures together in the optimisation problem and therefore can find their optimal combination for the globally best NN. Secondly, the computational cost can be reduced by using multi-scale accuracy evaluation methods to evaluate candidates. Thirdly, a stagnation-based switch strategy is proposed to adaptively switch different evaluation methods to better balance the search performance and computational cost. The SAFE-PSO algorithm is tested on two widely used datasets: The 10-category (i.e., CIFAR10) and the 100−category (i.e., CIFAR100). The experimental results show that SAFE-PSO is very effective and efficient, which can not only find a promising NN automatically but also find a better NN than compared algorithms at the same computational cost.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{nokey,

title = {A new differentiable architecture search method for optimizing convolutional neural networks in the digital twin of intelligent robotic grasping},

author = {Weifei Hu and jinyi Shao and Qing Jiao and Chuxuan Wang and Jin Cheng and Zhenyu Liu and Jianrong Tan},

url = {https://link.springer.com/article/10.1007/s10845-022-01971-8},

journal = { Journal of Intelligent Manufacturing},

keywords = {},

pubstate = {published},

tppubtype = {article}

}