Bandit-Driven Batch Selection for Robust Learning under Label Noise
This addresses label noise in real-world datasets, offering a scalable solution with a balanced trade-off between efficiency and efficacy, though it appears incremental as it builds on existing batch selection and bandit techniques.
The paper tackles robust learning under label noise by introducing a bandit-driven batch selection method for SGD, achieving superior performance on CIFAR-10 across various corruption levels without computational overhead from auxiliary models.
We introduce a novel approach for batch selection in Stochastic Gradient Descent (SGD) training, leveraging combinatorial bandit algorithms. Our methodology focuses on optimizing the learning process in the presence of label noise, a prevalent issue in real-world datasets. Experimental evaluations on the CIFAR-10 dataset reveal that our approach consistently outperforms existing methods across various levels of label corruption. Importantly, we achieve this superior performance without incurring the computational overhead commonly associated with auxiliary neural network models. This work presents a balanced trade-off between computational efficiency and model efficacy, offering a scalable solution for complex machine learning applications.