Engineering
a year ago

Optimal hyperparameters for image classification not always optimal, accuracy fluctuates.

International Journal of Advanced Computer Science and Applications
Method for Hyperparameter Tuning of EfficientNetV2-based Image Classification by Deliberately Modifying Optuna Tuned Result
Jin Shimazoe, Kohei Arai, Mari Oda

Paper Summary

Related papers

Theme background image
Engineering
a year ago

Revolutionizing Energy Efficiency: Xgboost Model Outperforms RNN in Load Monitoring

Electricity Forecasting
Demand Response
Water Distribution
Theme background image
Materials Science
2 years ago

Revolutionizing machine learning: Faster hyperparameter tuning boosts accuracy for organic molecule predictions

Materials Informatics
Mass Spectrometry
Learning with Noisy Labels
Theme background image
Computer Science
2 years ago

XGBoost Dominates Machine Learning Algorithms with Grid Search Optimization

Data Mining

A new method was developed to improve image classification using EfficientNetV2 by adjusting the hyperparameters found by Optuna. By changing the most important hyperparameters, the accuracy of classification improved significantly. Surprisingly, even changing the least important hyperparameters sometimes led to better accuracy than using the optimal settings. This shows that the best hyperparameters found by Optuna may not always be the most effective for image classification tasks.