TY - GEN
T1 - Identifying the Training Stop Point with Noisy Labeled Data
AU - Kamabattula, Sree Ram
AU - Devarajan, Venkat
AU - Namazi, Babak
AU - Sankaranarayanan, Ganesh
N1 - Publisher Copyright:
© 2020 IEEE.
PY - 2020/12
Y1 - 2020/12
N2 - Finding an early stopping point at maximum obtainable test accuracy (MOTA) is a challenging problem when training deep neural networks (DNNs) with noisy labeled data. Recent studies assume either that i) a clean validation set is available or ii) the noise ratio is known, or, both. However, often a clean validation set is unavailable, and the noise estimation can be inaccurate. To overcome these issues, we provide a novel training solution, free of these conditions. We analyze the rate of change of the training accuracy under different conditions to identify a training stop region. We further develop a heuristic algorithm (AutoTSP) to find a training stop point (TSP) at or close to MOTA. We validate the robustness of AutoTSP through several experiments on various datasets, noise ratios and architectures.
AB - Finding an early stopping point at maximum obtainable test accuracy (MOTA) is a challenging problem when training deep neural networks (DNNs) with noisy labeled data. Recent studies assume either that i) a clean validation set is available or ii) the noise ratio is known, or, both. However, often a clean validation set is unavailable, and the noise estimation can be inaccurate. To overcome these issues, we provide a novel training solution, free of these conditions. We analyze the rate of change of the training accuracy under different conditions to identify a training stop region. We further develop a heuristic algorithm (AutoTSP) to find a training stop point (TSP) at or close to MOTA. We validate the robustness of AutoTSP through several experiments on various datasets, noise ratios and architectures.
KW - MOTA
KW - Memorization stages
KW - TSP
UR - http://www.scopus.com/inward/record.url?scp=85113381549&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85113381549&partnerID=8YFLogxK
U2 - 10.1109/CSCI51800.2020.00084
DO - 10.1109/CSCI51800.2020.00084
M3 - Conference contribution
AN - SCOPUS:85113381549
T3 - Proceedings - 2020 International Conference on Computational Science and Computational Intelligence, CSCI 2020
SP - 457
EP - 463
BT - Proceedings - 2020 International Conference on Computational Science and Computational Intelligence, CSCI 2020
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2020 International Conference on Computational Science and Computational Intelligence, CSCI 2020
Y2 - 16 December 2020 through 18 December 2020
ER -