A Method of Deep Learning Model Optimization for Image Classification on Edge Device

Lee, Hyungkeuk; Lee, NamKyung; Lee, Sung‐Jin

doi:10.3390/s22197344

Cited by 10 publications

(10 citation statements)

References 57 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, this network compression is conducted in a manner that does not compromise the accuracy and latency values obtained in the previous stage. Based on the study by [ 5 ], network compression is performed through quantization and pruning to determine the lightweighted

and

. This task is carefully managed to minimize network size while maintaining the target accuracy, as it can potentially impact accuracy.…”

Section: System Modelmentioning

confidence: 99%

“…This task is carefully managed to minimize network size while maintaining the target accuracy, as it can potentially impact accuracy. Based on the research [ 5 ], we conduct network compression through quantization and pruning to determine the lightweighted

and

from

…”

Section: System Modelmentioning

confidence: 99%

“…However, to apply the information gathered from these various sensors to autonomous driving in real time, the corresponding calculations must be lightweight and accelerated [ 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 ]. Among these sensors, the tasks that require the highest computation and latency are 2D and 3D context-aware computations, which primarily involve cameras, lidar, and radar.…”

Section: Introductionmentioning

confidence: 99%

“…In studies [ 6 , 7 , 8 , 13 , 14 , 15 ], network weight reduction and acceleration efforts were conducted for camera-based 2D object detection and 2D segmentation calculations. In studies [ 5 , 14 , 15 ], quantization, pruning, and knowledge distillation methods for light weighting of deep learning were studied. Study [ 18 ] performs acceleration research for camera-based lane detection.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Optimal Configuration of Multi-Task Learning for Autonomous Driving

Jun,

Son,

Yoo

et al. 2023

Sensors

Self Cite

View full text Add to dashboard Cite

For autonomous driving, it is imperative to perform various high-computation image recognition tasks with high accuracy, utilizing diverse sensors to perceive the surrounding environment. Specifically, cameras are used to perform lane detection, object detection, and segmentation, and, in the absence of lidar, tasks extend to inferring 3D information through depth estimation, 3D object detection, 3D reconstruction, and SLAM. However, accurately processing all these image recognition operations in real-time for autonomous driving under constrained hardware conditions is practically unfeasible. In this study, considering the characteristics of image recognition tasks performed by these sensors and the given hardware conditions, we investigated MTL (multi-task learning), which enables parallel execution of various image recognition tasks to maximize their processing speed, accuracy, and memory efficiency. Particularly, this study analyzes the combinations of image recognition tasks for autonomous driving and proposes the MDO (multi-task decision and optimization) algorithm, consisting of three steps, as a means for optimization. In the initial step, a MTS (multi-task set) is selected to minimize overall latency while meeting minimum accuracy requirements. Subsequently, additional training of the shared backbone and individual subnets is conducted to enhance accuracy with the predefined MTS. Finally, both the shared backbone and each subnet undergo compression while maintaining the already secured accuracy and latency performance. The experimental results indicate that integrated accuracy performance is critically important in the configuration and optimization of MTL, and this integrated accuracy is determined by the ITC (inter-task correlation). The MDO algorithm was designed to consider these characteristics and construct multi-task sets with tasks that exhibit high ITC. Furthermore, the implementation of the proposed MDO algorithm, coupled with additional SSL (semi-supervised learning) based training, resulted in a significant performance enhancement. This advancement manifested as approximately a 12% increase in object detection mAP performance, a 15% improvement in lane detection accuracy, and a 27% reduction in latency, surpassing the results of previous three-task learning techniques like YOLOP and HybridNet.

show abstract

and

. This task is carefully managed to minimize network size while maintaining the target accuracy, as it can potentially impact accuracy.…”

Section: System Modelmentioning

confidence: 99%

and

from

…”

Section: System Modelmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Optimal Configuration of Multi-Task Learning for Autonomous Driving

Jun,

Son,

Yoo

et al. 2023

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…In particular, this topic covers sensing technologies necessary for supporting efficient data collection (Kwon and Seo 2022) and frameworks for virtual sensor configurations to collect data via multiple devices (Alberternst et al 2021). Learning technology (Topic 3) addresses the machine learning algorithms can be used in various AI service areas (e.g., smart city service and healthcare service), such as deep learning models for image classification and segmentation (Lee et al 2022b;Tseng et al 2021) and federated learning models for privacy protection of AI services (Rodríguez-Barroso et al 2020). Network quality (Topic 4) addresses the quality for data collection, analysis, and utilization for AI services.…”

Section: Key Topics (12) In the Ai Service Literaturementioning

confidence: 99%

Connecting artificial intelligence to value creation in services: mechanism and implications

Kim

2023

Serv Bus

View full text Add to dashboard Cite

Artificial intelligence (AI) is transforming services by providing personalized solutions, enhancing customer experience, and reducing operational costs. To tackle the challenges posed by the extensive and diverse literature on AI services, a comprehensive review was conducted using text mining techniques on journal articles. Twelve key research topics were identified, and the enabler–interface–business framework was developed. In addition, a value creation mechanism for AI services consisting of 6Cs (i.e., connection, collection, and computation, communication, control, and co-creation) was proposed. The study provides a complete overview of AI services, facilitating academic discussion and industrial transformation.

show abstract