Multiobjective Reinforcement Learning-Based Neural Architecture Search for Efficient Portrait Parsing

Lyu, Bo; Wen, Shiping; Shi, Kaibo; Huang, Tingwen

doi:10.1109/tcyb.2021.3104866

Cited by 34 publications

(6 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This idea is similar to the channel attention mechanism proposed in SENet [29]. Like CBAM [30], assuming that the input feature F has C channels with width W and height H, we can calculate the channel attention map M by formula (6):…”

Section: B Partial Channel Connections Based On Channel Attentionmentioning

confidence: 99%

See 1 more Smart Citation

Improved Differentiable Architecture Search With Multi-Stage Progressive Partial Channel Connections

Xue,

Lu,

Neri

et al. 2024

IEEE Trans. Emerg. Top. Comput. Intell.

View full text Add to dashboard Cite

Neural architecture search has attracted great atten-4 tion in the research community and has been successfully applied in 5 the industry recently. Differentiable architecture search (DARTS) 6 is an efficient architecture search method. However, the networks 7 searched by DARTS are often unstable due to the large gap in the 8 architecture depth between the search phase and the verification 9 phase. In addition, due to unfair exclusive competition between 10 different candidate operations, DARTS is prone to skip connection 11 aggregation, which may cause performance collapse. In this article, 12 we propose progressive partial channel connections based on chan-13 nel attention for differentiable architecture search (PA-DARTS) to 14 solve the above problems. In the early stage of searching, we only 15 select a few key channels for convolution using channel attention 16 and reserve all candidate operations. As the search progresses, 17 we gradually increase the number of channels and eliminate un-18 promising candidate operations to ensure that the search phase and 19 verification phase are all carried out on 20 cells. Due to the existence 20 of the partial channel connections based on channel attention, 21 we can eliminate the unfair competition between operations and 22 increase the stability of PA-DARTS. Experimental results showed 23 that PA-DARTS could achieve 97.59% and 83.61% classification 24 accuracy on CIFAR-10 and CIFAR-100, respectively. On ImageNet, 25 our algorithm achieved 75.3% classification accuracy. 26 Index Terms-Differentiable architecture search, exclusive 27 competition, channel attention, progressive partial channel 28 connections. 29 I. INTRODUCTION 30 I N RECENT years, deep neural networks have been widely 31 used in the industry due to their outstanding performance [1], 32 [2]. The topological structure of deep neural networks plays a 33 decisive role in their performance. Early neural network archi-34 tectures were mostly designed manually by professionals, such 35 as VGG [3], Resnet [4] and GoogleNet [5]. Although these net-36 works have performed impressively, designing an excellent deep 37 neural network architecture requires plenty of trial and error by 38 Manuscript

show abstract

Section: B Partial Channel Connections Based On Channel Attentionmentioning

confidence: 99%

“…In recent years, NAS has made great progress. Reinforcement learning (RL) [6] and evolutionary computing (EC) [7] are two widely used methods for NAS. The RL-based method regards NAS as the process of an agent's action.…”

mentioning

confidence: 99%

Improved Differentiable Architecture Search With Multi-Stage Progressive Partial Channel Connections

Xue,

Lu,

Neri

et al. 2024

IEEE Trans. Emerg. Top. Comput. Intell.

View full text Add to dashboard Cite

show abstract

“…Literature source [36] proposed the use of advanced artificial neural networks for quantitative modeling in economics. Literature sources [37][38] presented the latest research results of the current neural networks, although they have not been applied to the prediction of financial crises, which provide current financial crisis forecasting tools for a better reference.…”

Section: Related Knowledgementioning

confidence: 99%

Research on Financial Risk Crisis Prediction of Listed Companies Based on IWOA-BP Neural Network

Li¹,

Li²

2022

Journal of Internet Technology

View full text Add to dashboard Cite

<p>To avoid the risk brought by the financial crisis, the Improved Whale Optimization Algorithm-Back Propagation neural network (IWOA-BP) financial crisis early warning model is proposed. This paper selects the data from financial statements of some of the listed Chinese manufacturing companies from 2015-2019 as the research sample. First, the financial data of enterprises are screened by principal component analysis, and the early warning model is constructed from the financial and nonfinancial factors of six indicators: solvency, operating capacity, profitability, development capacity, cash flow and risk level factors. Second, the Whale Optimization Algorithm is optimized by the chaos strategy, as well as by the dynamic weight and sine cosine algorithm. Finally, the improved Whale Algorithm is optimized for BP neural network parameters. In the simulation experiments, the performance of the improved whale optimization algorithm is substantially improved. In addition, in the empirical analysis, compared to the prediction model with other algorithms, the prediction model of this paper has better results in terms of prediction accuracy.</p> <p> </p>

show abstract

“…With the development of the economy, the demand for electric energy is also increasing. In recent years, a lot of studies have made great efforts to propose an innovative deep‐learning model 4,5 and conduct a theoretical analysis 6–8 to improve robustness and flexibility 9 . Technological innovation, especially the innovation and improvement of electric load forecasting technology, is particularly important for promoting the development of the electric industry.…”

Section: Introductionmentioning

confidence: 99%

Short‐term load forecasting based on a generalized regression neural network optimized by an improved sparrow search algorithm using the empirical wavelet decomposition method

Fan¹,

Li²,

Zhang³

et al. 2023

Energy Science & Engineering

View full text Add to dashboard Cite

With the development of the electric market, electric load forecasting has been increasingly pursued by many scholars. Because the electric load is affected by many factors, it is characterized by volatility and uncertainty, and it cannot be forecasted accurately only by a single model. In the research, a short‐term load forecasting integrated model is proposed to solve the problem of inaccurate forecasting of a single model. The key point of using the integrated model to forecast is to optimize the decomposed sequence to improve the accuracy of the forecast. empirical wavelet decomposition (EWT) is used to decompose the sequence into stationary sequences and avoid modal aliasing; the sparrow search algorithm (SSA) simulates the forecasting and anti‐forecasting behavior of the sparrow population, which is very similar to the electricity consumption behavior of various industries and has good optimization effect; generalized regression neural network (GRNN) is used for forecast and reconstruction; This is the EWT‐SSA‐GRNN model. This paper studies and analyzes the power load of a city in southern Australia. The results show that the integrated model reduces volatility through decomposition and optimization, and can improve forecast accuracy.

show abstract

Multiobjective Reinforcement Learning-Based Neural Architecture Search for Efficient Portrait Parsing

Cited by 34 publications

References 29 publications

Improved Differentiable Architecture Search With Multi-Stage Progressive Partial Channel Connections

Improved Differentiable Architecture Search With Multi-Stage Progressive Partial Channel Connections

Research on Financial Risk Crisis Prediction of Listed Companies Based on IWOA-BP Neural Network

Short‐term load forecasting based on a generalized regression neural network optimized by an improved sparrow search algorithm using the empirical wavelet decomposition method

Contact Info

Product

Resources

About