Deep Learning for Imbalanced Data

We're interested in developing and improving deep learning techniques specifically designed to handle imbalanced datasets. Imbalanced data refers to datasets where the distribution of classes or categories is highly skewed, with one or a few classes being significantly underrepresented compared to others, which is the case in many of our applications, e.g. rare weather events, bees swarming or fraudulent interaction. We aim to address the challenges associated with imbalanced data, such as biased predictions and poor generalization, by developing novel algorithms, loss functions, sampling techniques, and model architectures that can effectively handle and mitigate the impact of class imbalance during training and inference in deep learning models.