When Deep Classifiers Agree: Analyzing Correlations between Learning Order and Image Statistics
This work provides insights into the fundamental learning dynamics of deep neural networks, which is incremental as it builds on prior empirical evidence about similarities in training processes.
The paper investigates the relationship between the order in which deep classifiers learn data instances and core image statistics, finding that agreement on learning order is independent of architectures, hyperparameters, or labels but correlates with dataset statistics across multiple datasets like CIFAR10 and ImageNet.
Although a plethora of architectural variants for deep classification has been introduced over time, recent works have found empirical evidence towards similarities in their training process. It has been hypothesized that neural networks converge not only to similar representations, but also exhibit a notion of empirical agreement on which data instances are learned first. Following in the latter works$'$ footsteps, we define a metric to quantify the relationship between such classification agreement over time, and posit that the agreement phenomenon can be mapped to core statistics of the investigated dataset. We empirically corroborate this hypothesis across the CIFAR10, Pascal, ImageNet and KTH-TIPS2 datasets. Our findings indicate that agreement seems to be independent of specific architectures, training hyper-parameters or labels, albeit follows an ordering according to image statistics.