Researchers have developed a new method called Sharpness-Guided Equilibrium Sampling (SGS) to improve the performance of models trained on long-tailed datasets. SGS dynamically adjusts the sampling probability of data points during training, prioritizing under-represented classes while down-weighting those that cause significant changes in the loss landscape when using Sharpness-Aware Minimization (SAM). This approach aims to achieve a more balanced and flatter loss landscape, leading to better generalization. SGS-SAM demonstrated significant improvements on CIFAR-100 LT and ImageNet-LT benchmarks, enhancing tail accuracy by up to 10.85 points and overall accuracy by 3.56 points, with only a marginal increase in training time. AI
IMPACT Enhances model generalization on datasets with skewed class distributions, crucial for real-world applications.
RANK_REASON Academic paper detailing a novel method for improving machine learning on imbalanced datasets. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
