Record link: 
http://hdl.handle.net/20.500.12708/77394
-
Title: 
Neural Network Compression Through Shunt Connections and Knowledge Distillation for Semantic Segmentation Problems
en
Citation: 
Haas, B., Wendt, A., Jantsch, A., & Wess, M. (2021). Neural Network Compression Through Shunt Connections and Knowledge Distillation for Semantic Segmentation Problems. In I. Maglogiannis, J. MacIntyre, & L. Iliadis (Eds.), Artificial Intelligence Applications and Innovations - 17th IFIP WG 12.5 International Conference, AIAI 2021, Hersonissos, Crete, Greece, June 25–27, 2021, Proceedings (pp. 349–361). Springer Nature Switzerland AG. https://doi.org/10.1007/978-3-030-79150-6_28
-
Publisher DOI: 
10.1007/978-3-030-79150-6_28
-
Publication Type: 
Inproceedings - Full-Paper Contribution
en
Language: 
English
-
Authors: 
Haas, Bernhard 
Wendt, Alexander 
Jantsch, Axel 
Wess, Matthias 
-
Organisational Unit: 
E384-02 - Forschungsbereich Systems on Chip 
-
Published in: 
Artificial Intelligence Applications and Innovations - 17th IFIP WG 12.5 International Conference, AIAI 2021, Hersonissos, Crete, Greece, June 25–27, 2021, Proceedings 
-
ISBN: 
978-3-030-79149-0
-
Volume: 
627
-
DOI of the book: 
10.1007/978-3-030-79150-6
-
Date (published): 
2021
-
Event name: 
Springer Nature 
-
Event place: 
Austria
-
Number of Pages: 
13
-
Publisher: 
Springer Nature Switzerland AG, Greece
-
Publisher: 
Springer, Cham
-
Peer reviewed: 
Yes
-
Keywords: 
Optimization; Accuracy; Machine learning; Shunt connections; Knowledge distillation; Latency; CIFAR; Cityscapes; DeepLab; MobileNet; Embedded machine learning
-
Abstract: 
Employing convolutional neural network models for large scale datasets represents a big challenge. Especially embedded devices with limited resources cannot run most state-of-the-art model architectures in real-time, necessary for many applications. This paper proves the applicability of shunt connections on large scale datasets and narrows this computational gap. Shunt connections is a proposed method for MobileNet compression. We are the first to provide results of shunt connections for the MobileNetV3 model and for segmentation tasks on the Cityscapes dataset, using the DeeplabV3 architecture, on which we achieve compression by 28%, while observing a 3.52 drop in mIoU. The training of shunt-inserted models are optimized through knowledge distillation. The full code used for this work will be available online.
en
Research Areas: 
Information Systems Engineering: 50%
Computer Engineering and Software-Intensive Systems: 50%
-
Science Branch: 
Elektrotechnik, Elektronik, Informationstechnik
-
Appears in Collections:
Conference Paper

Show full item record

Page view(s)

180
checked on Dec 1, 2023

Download(s)

4
checked on Dec 1, 2023

Google ScholarTM

Check