Downsampling Machine Learning, Deep learning
noun
Definition 1 (dataset-level reduction): The process of reducing the number of samples in a dataset or the resolution of a representation, typically to lower computational cost, compress information, or rebalance class distributions [IBM].
Example in context: “Specifically, they are created by downsampling the samples of each category to obey an exponential decay …” [Zhang et al. 2024].
Definition 2 (representation-level reduction): The process of reducing the spatial, temporal, or structural resolution of a representation, typically during feature extraction or pooling in deep learning models [IBM].
Example in context: “In the process of evaluating the important nodes, we use a downsampling strategy to perform a hierarchical pooling operation on the whole graph…” [Xie et al. 2024]
Synonym: down-sampling
Related terms: resampling, undersampling, pooling, subsampling