On Efficient Constructions of Checkpoints
@inproceedings{Chen2020OnEC, title={On Efficient Constructions of Checkpoints}, author={Y. Chen and Z. Liu and Bin Ren and Xin Jin}, booktitle={ICML}, year={2020} }
Efficient construction of checkpoints/snapshots is a critical tool for training and diagnosing deep learning models. In this paper, we propose a lossy compression scheme for checkpoint constructions (called LC-Checkpoint). LC-Checkpoint simultaneously maximizes the compression rate and optimizes the recovery speed, under the assumption that SGD is used to train the model. LC-Checkpointuses quantization and priority promotion to store the most crucial information for SGD to recover, and then… CONTINUE READING
Supplemental Presentations
Presentation Slides
One Citation
PipeSwitch: Fast Pipelined Context Switching for Deep Learning Applications
- Computer Science
- OSDI 2020
- 2020
- PDF
References
SHOWING 1-10 OF 38 REFERENCES
Potential benefits of delta encoding and data compression for HTTP
- Computer Science
- SIGCOMM '97
- 1997
- 407
- PDF
Fault Tolerance in Iterative-Convergent Machine Learning
- Computer Science, Mathematics
- ICML
- 2019
- 12
- Highly Influential
- PDF
Litz: Elastic Framework for High-Performance Distributed Machine Learning
- Computer Science
- USENIX Annual Technical Conference
- 2018
- 27
- PDF
Distributed GraphLab: A Framework for Machine Learning in the Cloud
- Computer Science
- Proc. VLDB Endow.
- 2012
- 571
- PDF
QSGD: Communication-Efficient SGD via Gradient Quantization and Encoding
- Computer Science
- NIPS
- 2017
- 445
- PDF
Deep Compression: Compressing Deep Neural Network with Pruning, Trained Quantization and Huffman Coding
- Computer Science
- ICLR
- 2016
- 4,046
- PDF
NeST: A Neural Network Synthesis Tool Based on a Grow-and-Prune Paradigm
- Computer Science
- IEEE Transactions on Computers
- 2019
- 91
- PDF