Complex hybrid weighted pruning method for accelerating convolutional neural networks

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: March 6, 2024

Abstract The increasing interest in filter pruning of convolutional neural networks stems from its inherent ability to effectively compress and accelerate these networks. Currently, is mainly divided into two schools: norm-based relation-based. These methods aim selectively remove the least important filters according predefined rules. However, limitations lie inadequate consideration diversity impact batch normalization (BN) layers on input next layer, which may lead performance degradation. To address above similarity-based methods, this study conducts empirical analyses reveal their drawbacks subsequently introduces a groundbreaking complex hybrid weighted method. By evaluating correlations norms between individual filters, as well parameters BN our method identifies prunes most redundant robust manner, thereby avoiding significant decreases network performance. We conducted comprehensive direct experiments different depths ResNet using publicly available image classification datasets, ImageNet CIFAR-10. results demonstrate efficacy approach. In particular, when applied ResNet-50 dataset, achieves reduction 53.5% floating-point operations, with loss only 0.6%.

Language: Английский

---

Enhancing the Performance of Machine Learning and Deep Learning-Based Flood Susceptibility Models by Integrating Grey Wolf Optimizer (GWO) Algorithm

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(14), P. 2595 - 2595

Published: July 16, 2024

Flooding is a recurrent hazard occurring worldwide, resulting in severe losses. The preparation of flood susceptibility map non-structural approach to management before its occurrence. With recent advances artificial intelligence, achieving high-accuracy model for mapping (FSM) challenging. Therefore, this study, various intelligence approaches have been utilized achieve optimal accuracy modeling address challenge. By incorporating the grey wolf optimizer (GWO) metaheuristic algorithm into models—including neural networks (RNNs), support vector regression (SVR), and extreme gradient boosting (XGBoost)—the objective generate maps evaluate variation performance. tropical Manimala River Basin India, severely battered by flooding past, has selected as test site. This 15 conditioning factors such aspect, enhanced built-up bareness index (EBBI), slope, elevation, geomorphology, normalized difference water (NDWI), plan curvature, profile soil adjusted vegetation (SAVI), stream density, texture, power (SPI), terrain ruggedness (TRI), land use/land cover (LULC) topographic wetness (TWI). Thus, six are produced applying RNN, SVR, XGBoost, RNN-GWO, SVR-GWO, XGBoost-GWO models. All models exhibited outstanding (AUC above 0.90) performance, performance ranks following order: RNN-GWO (AUC: 0.968) > 0.961) SVR-GWO 0.960) RNN 0.956) XGBoost 0.953) SVR 0.948). It was discovered that hybrid GWO optimization improved three RNN-GWO-based shows 8.05% MRB very susceptible floods. found SPI, LULC, TWI top five influential factors.

Language: Английский

---

Synergies and trade-offs of ecosystem services affected by land use structures of small watershed in the Loess Plateau

Journal of Environmental Management, Journal Year: 2023, Volume and Issue: 350, P. 119589 - 119589

Published: Nov. 29, 2023

Language: Английский

---

Energy-agriculture nexus: Exploring the future of artificial intelligence applications

Energy Nexus, Journal Year: 2023, Volume and Issue: 13, P. 100263 - 100263

Published: Dec. 24, 2023

Energy and agriculture are two independent sectors that share a mutual coexistence referred to as the energy-agriculture nexus. In an attempt facilitate capacity of this simultaneously, there is need for involvement latest technologies such artificial intelligence (AI). This research focused on incorporation AI along nexus, in explore applications, opportunities, challenges its potential implications various stakeholders. According intensive literature survey conducted, applications were found be significant rise since last decade, specifically prediction optimization respectively, focusing bioenergy (55%), energy use analysis (17%), process value chain (6%), energy-efficient irrigation greenhouse livestock management (2%), farm power machinery (4%) risk (4%). Challenges observed terms data availability, complexity heterogeneity, computing power, accountability transparency decision-making focus. order fully comprehend integration nexus develop strategies guidelines maximizing advantages technology while minimizing risks adverse effects stakeholders, more discussions recommended.

Language: Английский

---

Soil erosion and degradation assessment integrating multi-parametric methods of RUSLE model, RS, and GIS in the Shaqlawa agricultural area, Kurdistan Region, Iraq

Environmental Monitoring and Assessment, Journal Year: 2023, Volume and Issue: 195(10)

Published: Sept. 5, 2023

Language: Английский

---