Back to Search
Start Over
Toward the Predictability of Dynamic Real-Time DNN Inference
- Source :
- IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems. 41:2849-2862
- Publication Year :
- 2022
- Publisher :
- Institute of Electrical and Electronics Engineers (IEEE), 2022.
-
Abstract
- Deep neural networks (DNNs) have been widely used in many Cyber-Physical Systems (CPS). However, it is still challenging work to deploy DNNs in real-time systems. In particular, the execution time of DNN inference must be predictable, s.t. it could be known whether the run-time inference can complete within a required timing constraint. Moreover, the timing constraints may change dynamically with the run-time environment in many embedded applications, such as autonomous cars. A possible way to meet such dynamic real-time requirements is to execute different sub-networks of a DNN at run-time. However, improper construction of sub-networks may not only introduce unpredictable inference time, s.t. the real-timing constraints could be violated unexpectedly, but also has poor compatibility with the well-optimized machine learning framework (e.g., TensorFlow). In this paper, we study the predictability when executing different sub-networks of a DNN. In particular, we present a feature-wise run-time adaptation framework for DNN inference, which is implemented and validated on NVIDIA Jetson TX2 and Nano with TensorFlow. The experimental results show that our method can achieve predictable inference time in comparison with the state-of-the-art methods.
- Subjects :
- Computer science
business.industry
Inference
Machine learning
computer.software_genre
Computer Graphics and Computer-Aided Design
Execution time
Constraint (information theory)
Embedded applications
Deep neural networks
Artificial intelligence
Electrical and Electronic Engineering
Predictability
Adaptation (computer science)
business
computer
Software
Subjects
Details
- ISSN :
- 19374151 and 02780070
- Volume :
- 41
- Database :
- OpenAIRE
- Journal :
- IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
- Accession number :
- edsair.doi...........ca745977e9b4644c0c5e0240ab6f9313