One-shot Neural Architecture Search Research Articles

Balanced quantum neural architecture search

In the last decade, there has been continuously increasing attention on Neural Architecture Search (NAS). The design of network architecture is aimed at automatically generating efficient neural networks in the absence of prior knowledge. Most of current oneshot NAS methods are based on weight sharing. However, there are two main problems that lead to suboptimal search results. Firstly, the large scale supernet results in a complex and difficult search for optimal subnetwork structures in the large discrete search space. Secondly, weight sharing makes the internal weight parameters of the supernet cannot always vary towards the final weights required for the optimal network structure due to the influence of different structures. These problems are responsible for the fact that the performance of the subnetworks searched by the supernet cannot represent the real performance of the subnetworks trained from scratch. To solve these problems, we propose the method based on quantum evolution and balance pool, called BQNAS, which consists of two stages. The single path supernet is trained based on weight sharing and balance pool sampling method. Then, quantum parallelism is exploited for one-hot encoding and searching for the optimal subnetwork structure based on quantum evolutionary algorithms. It can greatly improve the search efficiency of oneshot approaches in subnetwork sampling and evaluate the real performance for searching for superior subnetwork structure. Extensive experiments on two datasets show that the proposed approach outperforms the state-of-the-art ones. Specifically, BQNAS achieves a top-1 accuracy of 97.27% on CIFAR-10 and 81.36% on CIFAR-100.

DIY Your EasyNAS for Vision: Convolution Operation Merging, Map Channel Reducing, and Search Space to Supernet Conversion Tooling.

Despite its popularity as a one-shot Neural Architecture Search (NAS) approach, the applicability of differentiable architecture search (DARTS) on complex vision tasks is still limited by the high computation and memory costs incurred by the over-parameterized supernet. We propose a new architecture search method called EasyNAS, whose memory and computational efficiency is achieved via our devised operator merging technique which shares and merges the weights of candidate convolution operations into a single convolution, and a dynamic channel refinement strategy. We also introduce a configurable search space-to-supernet conversion tool, leveraging the concept of atomic search components, to enable its application from classification to more complex vision tasks: detection and semantic segmentation. In classification, EasyNAS achieves state-of-the-art performance on the NAS-Bench-201 benchmark, attaining an impressive 76.2% accuracy on ImageNet. For detection, it achieves a mean average precision (mAP) of 40.1 with 120 frames per second (FPS) on MS-COCO test-dev. Additionally, we transfer the discovered architecture to the rotation detection task, where EasyNAS achieves a remarkable 77.05 mAP 50 on the DOTA-v1.0 test set, using only 21.1 M parameters. In semantic segmentation, it achieves a competitive mean intersection over union (mIoU) of 72.6% at 173 FPS on Cityscape, after searching for only 0.7 GPU-day.

PWSNAS: Powering Weight Sharing NAS With General Search Space Shrinking Framework.

Neural architecture search (NAS) depends heavily on an efficient and accurate performance estimator. To speed up the evaluation process, recent advances, like differentiable architecture search (DARTS) and One-Shot approaches, instead of training every model from scratch, train a weight-sharing super-network to reuse parameters among different candidates, in which all child models can be efficiently evaluated. Though these methods significantly boost search efficiency, they inherently suffer from inaccurate and unstable performance estimation. To this end, we propose a general and effective framework for powering weight-sharing NAS, namely, PWSNAS, by shrinking search space automatically, i.e., candidate operators will be discarded if they are less important. With the strategy, our approach can provide a promising search space of a smaller size by progressively simplifying the original search space, which can reduce difficulties for existing NAS methods to find superior architectures. In particular, we present two strategies to guide the shrinking process: detect redundant operators with a new angle-based metric and decrease the degree of weight sharing of a super-network by increasing parameters, which differentiates PWSNAS from existing shrinking methods. Comprehensive analysis experiments on NASBench-201 verify the superiority of our proposed metric over existing accuracy-based and magnitude-based metrics. PWSNAS can easily apply to the state-of-the-art NAS methods, e.g., single path one-shot neural architecture search (SPOS), FairNAS, ProxylessNAS, DARTS, and progressive DARTS (PDARTS). We evaluate PWSNAS and demonstrate consistent performance gains over baseline methods.

Adversarial Robustness in Graph-Based Neural Architecture Search for Edge AI Transportation Systems

Edge AI technologies have been used for many Intelligent Transportation Systems, such as road traffic monitor systems. Neural Architecture Search (NAS) is a typcial way to search high-performance models for edge devices with limited computing resources. However, NAS is also vulnerable to adversarial attacks. In this paper, A One-Shot NAS is employed to realize derivative models with different scales. In order to study the relation between adversarial robustness and model scales, a graph-based method is designed to select best sub models generated from One-Shot NAS. Besides, an evaluation method is proposed to assess robustness of deep learning models under various scales of models. Experimental results shows an interesting phenomenon about the correlations between network sizes and model robustness, reducing model parameters will increase model robustness under maximum adversarial attacks, while, increasing model paremters will increase model robustness under minimum adversarial attacks. The phenomenon is analyzed, that is able to help understand the adversarial robustness of models with different scales for edge AI transportation systems.

ATNAS: Automatic Termination for Neural Architecture Search

Neural architecture search (NAS) is a framework for automating the design process of a neural network structure. While the recent one-shot approaches have reduced the search cost, there still exists an inherent trade-off between cost and performance. It is important to appropriately stop the search and further reduce the high cost of NAS. Meanwhile, the differentiable architecture search (DARTS), a typical one-shot approach, is known to suffer from overfitting. Heuristic early-stopping strategies have been proposed to overcome such performance degradation. In this paper, we propose a more versatile and principled early-stopping criterion on the basis of the evaluation of a gap between expectation values of generalisation errors of the previous and current search steps with respect to the architecture parameters. The stopping threshold is automatically determined at each search epoch without cost. In numerical experiments, we demonstrate the effectiveness of the proposed method. We stop the one-shot NAS algorithms and evaluate the acquired architectures on the benchmark datasets: NAS-Bench-201 and NATS-Bench. Our algorithm is shown to reduce the cost of the search process while maintaining a high performance.

XploreNAS : Explore Adversarially Robust and Hardware-efficient Neural Architectures for Non-ideal Xbars

Compute In-Memory platforms such as memristive crossbars are gaining focus as they facilitate acceleration of Deep Neural Networks (DNNs) with high area and compute efficiencies. However, the intrinsic non-idealities associated with the analog nature of computing in crossbars limits the performance of the deployed DNNs. Furthermore, DNNs are shown to be vulnerable to adversarial attacks leading to severe security threats in their large-scale deployment. Thus, finding adversarially robust DNN architectures for non-ideal crossbars is critical to the safe and secure deployment of DNNs on the edge. This work proposes a two-phase algorithm-hardware co-optimization approach called XploreNAS that searches for hardware efficient and adversarially robust neural architectures for non-ideal crossbar platforms. We use the one-shot Neural Architecture Search approach to train a large Supernet with crossbar-awareness and sample adversarially robust Subnets therefrom, maintaining competitive hardware efficiency. Our experiments on crossbars with benchmark datasets (SVHN, CIFAR10, CIFAR100) show up to ~8–16% improvement in the adversarial robustness of the searched Subnets against a baseline ResNet-18 model subjected to crossbar-aware adversarial training. We benchmark our robust Subnets for Energy-Delay-Area-Products (EDAPs) using the Neurosim tool and find that with additional hardware efficiency–driven optimizations, the Subnets attain ~1.5–1.6× lower EDAPs than ResNet-18 baseline.

Tiny adversarial multi-objective one-shot neural architecture search

The widely employed tiny neural networks (TNNs) in mobile devices are vulnerable to adversarial attacks. However, more advanced research on the robustness of TNNs is highly in demand. This work focuses on improving the robustness of TNNs without sacrificing the model’s accuracy. To find the optimal trade-off networks in terms of the adversarial accuracy, clean accuracy, and model size, we present TAM-NAS, a tiny adversarial multi-objective one-shot network architecture search method. First, we build a novel search space comprised of new tiny blocks and channels to establish a balance between the model size and adversarial performance. Then, we demonstrate how the supernet facilitates the acquisition of the optimal subnet under white-box adversarial attacks, provided that the supernet significantly impacts the subnet’s performance. Concretely, we investigate a new adversarial training paradigm by evaluating the adversarial transferability, the width of the supernet, and the distinction between training subnets from scratch and fine-tuning. Finally, we undertake statistical analysis for the layer-wise combination of specific blocks and channels on the first non-dominated front, which can be utilized as a design guideline for the design of TNNs.

NAS-LID: Efficient Neural Architecture Search with Local Intrinsic Dimension

One-shot neural architecture search (NAS) substantially improves the search efficiency by training one supernet to estimate the performance of every possible child architecture (i.e., subnet). However, the inconsistency of characteristics among subnets incurs serious interference in the optimization, resulting in poor performance ranking correlation of subnets. Subsequent explorations decompose supernet weights via a particular criterion, e.g., gradient matching, to reduce the interference; yet they suffer from huge computational cost and low space separability. In this work, we propose a lightweight and effective local intrinsic dimension (LID)-based method NAS-LID. NAS-LID evaluates the geometrical properties of architectures by calculating the low-cost LID features layer-by-layer, and the similarity characterized by LID enjoys better separability compared with gradients, which thus effectively reduces the interference among subnets. Extensive experiments on NASBench-201 indicate that NAS-LID achieves superior performance with better efficiency. Specifically, compared to the gradient-driven method, NAS-LID can save up to 86% of GPU memory overhead when searching on NASBench-201. We also demonstrate the effectiveness of NAS-LID on ProxylessNAS and OFA spaces. Source code:https://github.com/marsggbo/NAS-LID.

RSBNet: One-shot neural architecture search for a backbone network in remote sensing image recognition

Recently, a massive number of deep learning-based approaches have been successfully applied to various remote sensing image (RSI) recognition tasks. However, most existing advances of deep learning methods in the RSI field heavily rely on the features extracted by the manually designed backbone network, which severely hinders the potential of deep learning models due to the complexity of RSI and the limitation of prior knowledge. In this paper, we research a new design paradigm for the backbone architecture in RSI recognition tasks, including scene classification, land-cover classification and object detection. A novel one-shot architecture search framework based on a weight-sharing strategy and an evolutionary algorithm is proposed, called RSBNet, which consists of three stages: Firstly, a supernet constructed in a layer-wise search space is pretrained on a self-assembled large-scale RSI dataset based on an ensemble single-path training strategy. Next, the pre-trained supernet is equipped with different recognition heads through the switchable recognition module and respectively fine-tuned on the target dataset to obtain task-specific supernet. Finally, we search for the optimal backbone architecture for different recognition tasks based on the evolutionary algorithm without any network training. Extensive experiments have been conducted on five benchmark datasets for different recognition tasks, the results show the effectiveness of the proposed search paradigm and demonstrate that the searched backbone is able to flexibly adapt different RSI recognition tasks and achieve impressive performance.

MNGNAS: Distilling Adaptive Combination of Multiple Searched Networks for One-Shot Neural Architecture Search.

Recently neural architecture (NAS) search has attracted great interest in academia and industry. It remains a challenging problem due to the huge search space and computational costs. Recent studies in NAS mainly focused on the usage of weight sharing to train a SuperNet once. However, the corresponding branch of each subnetwork is not guaranteed to be fully trained. It may not only incur huge computation costs but also affect the architecture ranking in the retraining procedure. We propose a multi-teacher-guided NAS, which proposes to use the adaptive ensemble and perturbation-aware knowledge distillation algorithm in the one-shot-based NAS algorithm. The optimization method aiming to find the optimal descent directions is used to obtain adaptive coefficients for the feature maps of the combined teacher model. Besides, we propose a specific knowledge distillation process for optimal architectures and perturbed ones in each searching process to learn better feature maps for later distillation procedures. Comprehensive experiments verify our approach is flexible and effective. We show improvement in precision and search efficiency in the standard recognition dataset. We also show improvement in correlation between the accuracy of the search algorithm and true accuracy by NAS benchmark datasets.

Optimized one-shot neural architecture search for skin cancer classification

Skin cancer is the world’s fifth most prevalent cancer. Skin cancer is seen in one out of every three cancers. The quality of early and precise skin cancer detection is quite difficult. Some studies have been conducted on the computerized detection of malignancy in skin lesion images. The manual tuning of network architecture design is heavily reliant on domain expertise and is tough. In the present research, a one-shot neural architecture search technique is proposed that boost the search efficiency by modeling NAS as a one-shot training process of an over-parameterized supernet. As a result, various architectures can be derived from the supernet, and share the same weights. An evolutionary memetic algorithm employing local search was proposed to optimum the network architecture, which yields minimal computation time and accurate classification of benign and malignant skin lesions. Our search algorithm comprises all conceivable networks in each solution space. This approach prevents the neural architecture from being scratched. The proposed work proved efficient in ResNet search space with better outcomes than manually created networks on the ISBI dataset, which resulted in an accuracy of 93.26% and the computation cost is decreased. The search process also lasts an average of 18 h on a graphical processing unit.

Evolutionary Search for Complete Neural Network Architectures With Partial Weight Sharing

Neural architecture search (NAS) provides an automatic solution in designing network architectures. Unfortunately, the direct search for complete task-dependent network architectures is laborious since training and evaluating complete neural architectures over a large search space are computationally prohibitive. Recently, one-shot NAS (OSNAS) has attracted great attention in the NAS community because it significantly speeds up the candidate architecture evaluation procedure through weight sharing. However, the full weight sharing training paradigm in OSNAS may result in strong interference across candidate architectures and mislead the architecture search. To alleviate the problem, we propose a partial weight sharing OSNAS framework that directly evolves complete neural network architectures. In particular, we suggest a novel node representation scheme that randomly activates a subset of nodes of the one-shot model in each generation to reduce the weight coupling in the one-shot model. During the evolutionary search, a tailored crossover operator randomly samples the nodes from two parent individuals or a single parent to construct new candidate architectures, thus effectively constraining the degree of weight sharing. Furthermore, we introduce a new mutation operator that replaces the chosen nodes of the one-shot model with randomly generated nodes to enhance the exploratory capability. Finally, we encode a set of pyramidal convolution operations in the search space, enabling the evolved neural networks to capture different levels of details in the images. The proposed method is examined and compared with 26 state-of-the-art algorithms on ten image classification tasks, including CIFAR series, CINIC10, ImageNet, and MedMNIST series. The experimental results demonstrate that the proposed method can computationally much more efficiently find neural architectures that achieve comparable classification accuracy to the state-of-the-art designs.

One-shot neural architecture search for fault diagnosis using vibration signals

Machine learning method has been widely applied in industrial fault diagnosis, especially the deep learning method. In the field of industrial fault diagnosis, deep learning is mostly used to extract features of vibration signals to achieve end-to-end fault diagnosis systems. Due to the complexity and variety of actual industrial datasets, some deep learning models are designed to be complicated. However, designing neural network architectures requires rich professional knowledge, experience, and a large number of experiments, increasing the difficulty of developing deep learning models. Fortunately, Neural Architecture Search (NAS), a branch of Automated Machine Learning (AutoML), is developing rapidly. Given a search space, NAS can search for networks that perform better than manually designed. In this paper, a one-shot NAS method for fault diagnosis is proposed. The one-shot model is a supernet that contains all candidate networks in a given search space. The supernet is trained to evaluate the actual performance of candidate networks by measuring the difference between its output probability and the true labels. According to the prediction of supernet, the networks with excellent performance can be searched, using some common search methods such as random search or evolutionary algorithm. Finally, the searched network is trained by reusing the weights of the supernet. To evaluate the proposed method, two search spaces are designed, ResNet and Inception search spaces, to search on PHM 2009 Data Challenge gearbox dataset. The state-of-the-art results are obtained, and accuracies of searched ResNet-A and Inception-A are 84.11% and 83.81%, which are 3.29% and 10.88% higher than Reinforcement Learning based NAS.

One-Shot Neural Architecture Search by Dynamically Pruning Supernet in Hierarchical Order.

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.

Rethinking Bi-Level Optimization in Neural Architecture Search: A Gibbs Sampling Perspective

One-Shot architecture search, which aims to explore all possible operations jointly based on a single model, has been an active direction of Neural Architecture Search (NAS). As a well-known one-shot solution, Differentiable Architecture Search (DARTS) performs continuous relaxation on the architecture's importance and results in a bi-level optimization problem. However, as many recent studies have shown, DARTS cannot always work robustly for new tasks, which is mainly due to the approximate solution of the bi-level optimization. In this paper, one-shot neural architecture search is addressed by adopting a directed probabilistic graphical model to represent the joint probability distribution over data and model. Then, neural architectures are searched for and optimized by Gibbs sampling. We rethink the bi-level optimization problem as the task of Gibbs sampling from the posterior distribution, which expresses the preferences for different models given the observed dataset. We evaluate our proposed NAS method -- GibbsNAS on the search space used in DARTS/ENAS and the search space of NAS-Bench-201. Experimental results on multiple search space show the efficacy and stability of our approach.

AdvantageNAS: Efficient Neural Architecture Search with Credit Assignment

Neural architecture search (NAS) is an approach for automatically designing a neural network architecture without human effort or expert knowledge. However, the high computational cost of NAS limits its use in commercial applications. Two recent NAS paradigms, namely one-shot and sparse propagation, which reduce the time and space complexities, respectively, provide clues for solving this problem. In this paper, we propose a novel search strategy for one-shot and sparse propagation NAS, namely AdvantageNAS, which further reduces the time complexity of NAS by reducing the number of search iterations. AdvantageNAS is a gradient-based approach that improves the search efficiency by introducing credit assignment in gradient estimation for architecture updates. Experiments on the NAS-Bench-201 and PTB dataset show that AdvantageNAS discovers an architecture with higher performance under a limited time budget compared to existing sparse propagation NAS. To further reveal the reliabilities of AdvantageNAS, we investigate it theoretically and find that it monotonically improves the expected loss and thus converges.

Improving One-Shot NAS with Shrinking-and-Expanding Supernet

Training a supernet using a copy of shared weights has become a popular approach to speed up neural architecture search (NAS). However, it is difficult for supernet to accurately evaluate on a large-scale search space due to high weight coupling in weight-sharing setting. To address this, we present a shrinking-and-expanding supernet that decouples the shared parameters by reducing the degree of weight sharing, avoiding unstable and inaccurate performance estimation as in previous methods. Specifically, we propose a new shrinking strategy that progressively simplifies the original search space by discarding unpromising operators in a smart way. Based on this, we further present an expanding strategy by appropriately increasing parameters of the shrunk supernet. We provide comprehensive evidences showing that, in weight-sharing supernet, the proposed method SE-NAS brings more accurate and more stable performance estimation. Experimental results on ImageNet dataset indicate that SE-NAS achieves higher Top-1 accuracy than its counterparts under the same complexity constraint and search space. The ablation study is presented to further understand SE-NAS.

Efficient pyramid context encoding and feature embedding for semantic segmentation

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.

Bayesian Differentiable Architecture Search for Efficient Domain Matching Fault Diagnosis

Intelligent fault diagnosis is essential for downtime reduction in modern industries. However, domain (i.e., working condition) variants are unavoidable for fault diagnosis under changing environment. Though one can obtain a customized deep learning model for a given domain, it is not practical to design deep neural networks manually for multiple domains. Differentiable architecture search (DARTS), as an automatic machine learning technique, can automate the network design process for a specific domain efficiently by using hypernetwork and differentiable search strategy. Nevertheless, the representation of hypernetwork is restricted by several unfair factors, and the searched architecture of DARTS is overconfident. To address these issues, a one-shot neural architecture search approach, which involves two-stage learning, is proposed for efficient domain matching fault diagnosis. In the first training stage, the warmup and path-dropout strategies are taken to enhance the competitiveness of the parametric operators and alleviate the coadaptation problem to obtain an intuitively fair hypernetwork. In the second matching stage, variational inference is introduced into a differentiable search strategy to estimate the uncertainty of model matching, and a scale mixture prior is used to softly constrain the matching stage. Then, a candidate architectures set can be sampled and ordered from the posterior. Multidomain experiment is implemented by adding noise to the raw signal, and the proposed method outperforms four commonly used deep neural networks for aeroengine bevel gear fault diagnosis.

Event Classification with Multi-step Machine Learning

The usefulness and value of Multi-step Machine Learning (ML), where a task is organized into connected sub-tasks with known intermediate inference goals, as opposed to a single large model learned end-to-end without intermediate sub-tasks, is presented. Pre-optimized ML models are connected and better performance is obtained by re-optimizing the connected one. The selection of an ML model from several small ML model candidates for each sub-task has been performed by using the idea based on Neural Architecture Search (NAS). In this paper, Differentiable Architecture Search (DARTS) and Single Path One-Shot NAS (SPOS-NAS) are tested, where the construction of loss functions is improved to keep all ML models smoothly learning. Using DARTS and SPOS-NAS as an optimization and selection as well as the connections for multi-step machine learning systems, we find that (1) such a system can quickly and successfully select highly performant model combinations, and (2) the selected models are consistent with baseline algorithms, such as grid search, and their outputs are well controlled.