TY - GEN
T1 - Coreset-based neural network compression
AU - Dubey, Abhimanyu
AU - Chatterjee, Moitreya
AU - Ahuja, Narendra
N1 - Funding Information:
We are grateful to Prof. Ramesh Raskar for his insightful comments. MC additionally acknowledges Po-han Huang for helpful discussions and NVIDIA for providing the GPUs used for this research.
Publisher Copyright:
© Springer Nature Switzerland AG 2018.
PY - 2018
Y1 - 2018
N2 - We propose a novel Convolutional Neural Network (CNN) compression algorithm based on coreset representations of filters. We exploit the redundancies extant in the space of CNN weights and neuronal activations (across samples) in order to obtain compression. Our method requires no retraining, is easy to implement, and obtains state-of-the-art compression performance across a wide variety of CNN architectures. Coupled with quantization and Huffman coding, we create networks that provide AlexNet-like accuracy, with a memory footprint that is 832 × smaller than the original AlexNet, while also introducing significant reductions in inference time as well. Additionally these compressed networks when fine-tuned, successfully generalize to other domains as well.
AB - We propose a novel Convolutional Neural Network (CNN) compression algorithm based on coreset representations of filters. We exploit the redundancies extant in the space of CNN weights and neuronal activations (across samples) in order to obtain compression. Our method requires no retraining, is easy to implement, and obtains state-of-the-art compression performance across a wide variety of CNN architectures. Coupled with quantization and Huffman coding, we create networks that provide AlexNet-like accuracy, with a memory footprint that is 832 × smaller than the original AlexNet, while also introducing significant reductions in inference time as well. Additionally these compressed networks when fine-tuned, successfully generalize to other domains as well.
UR - http://www.scopus.com/inward/record.url?scp=85055091388&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85055091388&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-01234-2_28
DO - 10.1007/978-3-030-01234-2_28
M3 - Conference contribution
AN - SCOPUS:85055091388
SN - 9783030012335
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 469
EP - 486
BT - Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
A2 - Ferrari, Vittorio
A2 - Sminchisescu, Cristian
A2 - Hebert, Martial
A2 - Weiss, Yair
PB - Springer
T2 - 15th European Conference on Computer Vision, ECCV 2018
Y2 - 8 September 2018 through 14 September 2018
ER -