Measurement, Journal Year: 2025, Volume and Issue: unknown, P. 117761 - 117761
Published: May 1, 2025
Language: Английский
Measurement, Journal Year: 2025, Volume and Issue: unknown, P. 117761 - 117761
Published: May 1, 2025
Language: Английский
AppliedMath, Journal Year: 2025, Volume and Issue: 5(1), P. 8 - 8
Published: Jan. 14, 2025
Optimizing learning rates (LRs) in deep (DL) has long been challenging. Previous solutions, such as rate scheduling (LRS) and adaptive (ALR) algorithms like RMSProp Adam, added complexity by introducing new hyperparameters, thereby increasing the cost of model training through expensive cross-validation experiments. These methods mainly focus on local gradient patterns, which may not be effective scenarios with multiple optima near global optimum. A technique called Learning Rate Tuner Relative Adaptation (LRT-RA) is introduced to tackle these issues. This approach dynamically adjusts LRs during analyzing loss curve, eliminating need for costly initial LR estimation cross-validation. method reduces expenses carbon footprint enhances efficiency. It demonstrates promising results preventing premature convergence, exhibiting inherent optimization behavior, elucidating correlation between dataset distribution optimal selection. The proposed achieves 84.96% accuracy CIFAR-10 while reducing power usage 0.07 kWh, CO2 emissions 0.05, both SO2 NOx 0.00003 pounds, whole testing process.
Language: Английский
Citations
0Advanced Engineering Informatics, Journal Year: 2025, Volume and Issue: 65, P. 103211 - 103211
Published: March 3, 2025
Language: Английский
Citations
0Expert Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 127715 - 127715
Published: April 1, 2025
Citations
0Communications in Nonlinear Science and Numerical Simulation, Journal Year: 2025, Volume and Issue: unknown, P. 108904 - 108904
Published: April 1, 2025
Language: Английский
Citations
0Measurement, Journal Year: 2025, Volume and Issue: unknown, P. 117737 - 117737
Published: May 1, 2025
Language: Английский
Citations
0Measurement, Journal Year: 2025, Volume and Issue: unknown, P. 117761 - 117761
Published: May 1, 2025
Language: Английский
Citations
0