Filters








3,775 Hits in 3.5 sec

Adaptable mobile vision systems through multi-exit neural networks

Alexandros Kouris, Stylianos I. Venieris, Stefanos Laskaridis, Nicholas D. Lane
2022 Proceedings of the 20th Annual International Conference on Mobile Systems, Applications and Services  
Semantic segmentation constitutes the backbone of many mobile vision systems, spanning from robot navigation to augmented reality and teleconferencing.  ...  Upon deployment, the predictions of these exits can be exploited either in a dynamic (input-adaptive) way, to save computation during inference on easier samples; or in a static (device-adaptive) setting  ...  Figure 1 : 1 Figure 1: Multi-exit segmentation network instance. Figure 2 : 2 Figure 2: Progressive segmentation through MESS.  ... 
doi:10.1145/3498361.3538791 fatcat:udj2bntzj5ekbl7x2cdd5maudm

Boosting Mobile CNN Inference through Semantic Memory [article]

Yun Li, Chen Zhang, Shihao Han, Li Lyna Zhang, Baoqun Yin, Yunxin Liu, Mengwei Xu
2021 arXiv   pre-print
SMTM is prototyped on commodity CNN engine and runs on both mobile CPU and GPU.  ...  timing considering different layers' inherent characteristics; (3) it adaptively adjusts the cache size and semantic vectors to fit the scene dynamics.  ...  Multi-branch neural architectures.  ... 
arXiv:2112.02644v1 fatcat:gfyecsojvzgaxjgmlwihov26ju

面向实时视频流分析的边缘计算技术

Zheng Yang, Xiaowu He, Jiaxing Wu, Xu Wang, Yi Zhao
2021 Scientia Sinica Informationis  
ImageNet classification with deep convolutional neural networks. In: Proceedings of Advances in Neural Information Processing Systems (NeurIPS). Lake Tahoe, NV, USA: Curran Associates, Inc., 2011.  ...  Large-scale video classification with convolutional neural networks. In: Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).  ... 
doi:10.1360/ssi-2021-0133 fatcat:qs7jnvnknjhdrhfrru6rfbwuge

A Survey on Collaborative DNN Inference for Edge Intelligence [article]

Weiqing Ren, Yuben Qu, Chao Dong, Yuqian Jing, Hao Sun, Qihui Wu, Song Guo
2022 arXiv   pre-print
With the vigorous development of artificial intelligence (AI), the intelligent applications based on deep neural network (DNN) change people's lifestyles and the production efficiency.  ...  To solve the above problems, by embedding AI model training and inference capabilities into the network edge, edge intelligence (EI) becomes a cutting-edge direction in the field of AI.  ...  While being able to adapt to the DNN inference on cloud, DDNN also allows fast and local inference using the shallow part of the neural network on edge and terminal devices.  ... 
arXiv:2207.07812v1 fatcat:yahjwuowz5erhctsgbderya65m

Multi-person location and tracking method based on BP neural network

Pan Wei, Liu Zhizhan, Zou Yi
2008 2008 IEEE Conference on Cybernetics and Intelligent Systems  
To establish the complicated relationship between the 2D-image information that is obtained through the three-camera system and the 3D information of the target, an artificial neural network is proposed  ...  Experiment results verify that the BP neural network improves the efficiency, accuracy and robustness of the method comparing with Traditional Binocular Location (TBL) methods.  ...  Figure 1 . 1 Figure 2 . 112 Three-Vision cameras system (3D) Three-Vision cameras system (2D) Different from Traditional Binocular Location (TBL) methods based on neural networks, this paper use the 3ds  ... 
doi:10.1109/iccis.2008.4670911 fatcat:kx4dvhv5cfd37nytqrtwm3zkfu

Multi-Exit Vision Transformer for Dynamic Inference [article]

Arian Bakhtiarnia, Qi Zhang, Alexandros Iosifidis
2021 arXiv   pre-print
Deep neural networks can be converted to multi-exit architectures by inserting early exit branches after some of their intermediate layers.  ...  In particular, in edge computing systems and IoT networks where the exact computation time budget is variable and not known beforehand.  ...  To the best of our knowledge, multi-exit Vision Transformer architectures have not yet been studied in the literature, which limits the application of Vision Transformers in mobile and edge computing.  ... 
arXiv:2106.15183v3 fatcat:pe5yulnrzjbnfopwosvaw443am

A Wireless Actuator-Sensor Neural Network for Evacuation Routing

A. Jankowska, M. C. Schut, N. Ferreira-Schut
2009 2009 Third International Conference on Sensor Technologies and Applications  
We explain how the combination of sensor network and neural network technology can describe a system which maintains desired characteristics such as scalability and adaptability.  ...  In this paper, we look at the application of neural networks to individual nodes in a wireless network, which result in a wireless sensor-actuator neural network model.  ...  In Section 2, we present background material concerning the combination of multi-agent systems and WSANs, the use of neural networks in WSANs, and the use WS(A)Ns for evacuation routing.  ... 
doi:10.1109/sensorcomm.2009.30 fatcat:mzntbfzgwbehpfplz3s3m7ztfq

Fluid Batching: Exit-Aware Preemptive Serving of Early-Exit Neural Networks on Edge NPUs [article]

Alexandros Kouris, Stylianos I. Venieris, Stefanos Laskaridis, Nicholas D. Lane
2022 arXiv   pre-print
With deep neural networks (DNNs) emerging as the backbone in a multitude of computer vision tasks, their adoption in real-world consumer applications broadens continuously.  ...  At the same time, the deployment of dynamic DNNs, comprising stochastic computation graphs (e.g. early-exit (EE) models), introduces a new dimension of dynamic behaviour in such systems.  ...  Early-Exit Neural Networks Dynamic DNNs come in many shapes and forms [8] - [11] .  ... 
arXiv:2209.13443v1 fatcat:iue4hsh4kjfhfmgul3biomo2ji

Split Computing and Early Exiting for Deep Learning Applications: Survey and Research Challenges [article]

Yoshitomo Matsubara, Marco Levorato, Francesco Restuccia
2022 arXiv   pre-print
Mobile devices such as smartphones and autonomous vehicles increasingly rely on deep neural networks (DNNs) to execute complex inference tasks such as image classification and speech recognition, among  ...  Although task offloading to cloud/edge servers may decrease the mobile device's computational burden, erratic patterns in channel quality, network, and edge server load can lead to a significant delay  ...  [97] propose a supervised compression method for resource-constrained edge computing systems, which adapts ideas from knowledge distillation and neural image compression [4, 5] .  ... 
arXiv:2103.04505v4 fatcat:2qfyc4o4mfglhftw4noccupmzm

Multi-Exit Vision Transformer for Dynamic Inference [article]

Arian Bakhtiarnia, Qi Zhang, Alexandros Iosifidis
2021 Zenodo  
Deep neural networks can be converted to multi- exit architectures by inserting early exit branches after some of their intermediate layers.  ...  In particular, in edge computing systems and IoT networks where the exact computa- tion time budget is variable and not known beforehand.  ...  To the best of our knowledge, multi-exit Vision Transformer architectures have not yet been studied in the literature, which limits the application of Vision Transformers in mobile and edge computing.  ... 
doi:10.5281/zenodo.5604478 fatcat:q3rxb4zq5zbwhiuavttlcvp3oy

Selective Fine-Tuning on a Classifier Ensemble: Realizing Adaptive Neural Networks with a Diversified Multi-Exit Architecture

Hirose Kazutoshi, Shinya Takamaeda-Yamazaki, Jaehoon Yu, Masato Motomura
2020 IEEE Access  
However, a multi-exit network, which realizes adaptive inference costs, requires significant training costs because it has many classifiers that need to be fine-tuned.  ...  In this study, we propose a novel fine-tuning method for an ensemble of classifiers that efficiently retrain the multi-exit network.  ...  We will show the effectiveness of exploiting the individuality of each classifier in a multi-exit network through the experimental results. A.  ... 
doi:10.1109/access.2020.3047799 fatcat:zgqebi3yxneb7aeaymlcsccqxi

MobiVQA

Qingqing Cao, Prerna Khanna, Nicholas D. Lane, Aruna Balasubramanian
2022 Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies  
Existing mobile optimizations that work for vision-only or text-only neural networks cannot be applied here because of the dependencies between the two modes.  ...  Unfortunately, existing techniques that optimize deep learning for mobile devices cannot be applied for VQA because the VQA task is multi-modal---it requires both processing vision and text data.  ...  Early Exit. BranchyNet [53] , DenseNet [28] , and Shallow-Deep Network [32] are pioneering works in the computer vision community that started the early exit idea for efficient neural network.  ... 
doi:10.1145/3534619 fatcat:malh4ljosjdixddfpjlofah6v4

Dynamic Neural Networks: A Survey [article]

Yizeng Han, Gao Huang, Shiji Song, Le Yang, Honghui Wang, Yulin Wang
2021 arXiv   pre-print
Dynamic neural network is an emerging research topic in deep learning.  ...  Compared to static models which have fixed computational graphs and parameters at the inference stage, dynamic networks can adapt their structures or parameters to different inputs, leading to notable  ...  Training objectives for efficient inference 1) Training multi-exit networks.  ... 
arXiv:2102.04906v4 fatcat:zelspxwv6nel7kv2yu6ynakyuu

A Survey of Intelligent Car Parking System

Faheem, S.A. Mahmud, G.M. Khan, M. Rahman, H. Zafar
2013 Journal of Applied Research and Technology  
Intelligent Parking Service is a part of Intelligent Transportation Systems (ITS).  ...  The discussed systems will be able to reduce the problems which are arising due to unavailability of a reliable, efficient and modern parking system, while the economic analysis technique will help in  ...  For outdoor mobile robot navigation, a multi-agent system with eventdriven control approach was adapted in [18] .  ... 
doi:10.1016/s1665-6423(13)71580-3 fatcat:3m3t3icyb5amzcsxl3xenyudvi

Study Of Deep Learning Techniques For Differently Abled Applications And Detection Techniques

Anandh N, Et. al.
2021 Turkish Journal of Computer and Mathematics Education  
Vision is the major sensing organ of the human and to assist the VIP regarding this, there are various digital vision products are in the market which is based on digital technologies and advanced algorithms  ...  This will transform the VIP's vision world into audio to get to know about their surroundings includes objects, motion, obstacles and spatial locations.  ...  They developed FR systems using cloud and fog computing with deep learning approach called deep convolutional neural networks (DCNN).  ... 
doi:10.17762/turcomat.v12i10.5396 fatcat:zw45ghte4vaavpjgdxa4emjlb4
« Previous Showing results 1 — 15 out of 3,775 results