Abstract: The success of deep neural networks (DNNs) is attributable to three factors:increased compute capacity, more complex models, and more data. These factors,however, are not always present, especially for edge applications such asautonomous driving, augmented reality, and internet-of-things. Training DNNsrequires a large amount of data, which is difficult to obtain. Edge devicessuch as mobile phones have limited compute capacity, and therefore, requirespecialized and efficient DNNs. However, due to the enormous design space andprohibitive training costs, designing efficient DNNs for different targetdevices is challenging. So the question is, with limited data, computecapacity, and model complexity, can we still successfully apply deep neuralnetworks?
This dissertation focuses on the above problems and improving the efficiencyof deep neural networks at four levels. Model efficiency: we designed neuralnetworks for various computer vision tasks and achieved more than 10x fasterspeed and lower energy. Data efficiency: we developed an advanced tool thatenables 6.2x faster annotation of a LiDAR point cloud. We also leveraged domainadaptation to utilize simulated data, bypassing the need for real data.Hardware efficiency: we co-designed neural networks and hardware acceleratorsand achieved 11.6x faster inference. Design efficiency: the process of findingthe optimal neural networks is time-consuming. Our automated neuralarchitecture search algorithms discovered, using 421x lower computational costthan previous search methods, models with state-of-the-art accuracy andefficiency.
Submission history
From: Bichen Wu [view email]Buy Efficient Networks SpeedStream SS1021 - Network adapter - PC Card - 802.11b: Network Cards - Amazon.com FREE DELIVERY possible on eligible purchases. A network g ∈ G is efficient if ϒ g ⩾ ϒ g ′, ∀ g ′ ∈ G. A network value is a common value that may be distributed among players, wireless nodes in our case. This is done by means of an allocation rule. Definition 4 Allocation Rule. This item Siemens Efficient Networks SpeedStream 5100 - DSL Modem (12A-5100-701) Actiontec GT701D Ethernet DSL Modem with Routing Capabilities NETGEAR High-Speed Broadband DSL Modem (DM200-100NAS). With a lot of cognitive network-based EE applications proposed in literature, artificial intelligence is expected to play a crucial role in EE for future networks, including for efficient adaptive resource allocation, discontinuous reception, channel learning for power management 76, 77, traffic offloading for energy efficiency in small.
[v1]Tue, 20 Aug 2019 23:26:04 UTC (21,014 KB)
Full-text links:
Download:
Current browse context:References & Citations
DBLP - CS Bibliography
Efficient Networks 5851 Manual
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs and how to get involved.