This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
Uncertainty-Aware Generative Oversampling Using an Entropy-Guided Conditional Variational Autoencoder
Created by
Haebom
Author
Amirhossein Zare (SeyedAbolfazl), Amirhessam Zare (SeyedAbolfazl), Parmida Sadat Pezeshki (SeyedAbolfazl), Herlock (SeyedAbolfazl), Rahimi, Ali Ebrahimi, Ignacio Vazquez -Garcia , Leo Anthony Celi
Outline
Class imbalance is a critical challenge in high-dimensional biomedical data, and we propose LEO-CVAE. LEO-CVAE is a generative oversampling framework that explicitly integrates local uncertainty into representation learning and data generation. It quantifies uncertainty using Shannon entropy, emphasizes robust learning in uncertain regions through local entropy-weighted loss (LEWL), and focuses generation on informative and class-overlapping regions using an entropy-based sampling strategy. We apply it to a clinical genomics dataset and demonstrate performance improvements over existing oversampling and generative baselines.
Takeaways, Limitations
•
Takeaways:
◦
Proposing a generative oversampling framework through uncertainty awareness.
◦
Effective in solving class imbalance problems.
◦
Demonstrated performance improvement by applying it to clinical genomics data.
◦
Suitable for areas with complex nonlinear structures.
•
Limitations:
◦
Based on CVAE, model complexity and computational cost may be high.
◦
Possible performance bias for specific datasets.
◦
There is room for improvement in how uncertainty is measured and utilized.
