Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Olshenbaum, Konstantin
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2025
Schlagworte:
Online-Zugang:https://doi.org/10.5281/zenodo.15948128
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Inhaltsangabe:
  • <p><strong>BoostNet-WB</strong> is a practical modification of the AdaBoost algorithm that integrates compact neural networks as weak learners. Instead of classical reweighting, the method employs a weighted bootstrap resampling strategy to reflect example importance during stochastic gradient descent — without modifying standard loss functions or optimizers.</p> <p>The approach includes dynamically adjustable network architectures and regularization techniques to prevent overfitting while preserving weak learner behavior. Experimental results across synthetic datasets show that BoostNet-WB consistently outperforms classical decision stump-based AdaBoost in classification performance, and the adaptive version achieves a favorable trade-off between accuracy and training time.</p> <p>This work highlights the feasibility of neural networks in boosting frameworks and offers a foundation for future extensions, including deeper or convolutional architectures.</p>