Fig. 10From: Multi-sample \(\zeta \)-mixup: richer, more realistic synthetic samples from a p-series interpolantHyperparameter sensitivity analysis for \(\zeta \)-mixup  on CIFAR-10 and CIFAR-100. In a, b, \(\gamma \) is varied from [1.8, 5.0] and the resulting ERR is shown. In c, d, 200 models are trained by varying \(\gamma \) uniformly in [1.0, 6.0] and weight decay log-uniformly in [5e\(-5, 1\)e\(-3]\). Each model is denoted by a curved line passing through the value of \(\gamma \) (left column) and weight decay (middle column) used for training, connecting it to the corresponding model’s test accuracy (right column). The lines are color-coded according to the models’ test accuracy. Models with \(\gamma < \gamma _{\textrm{min}}\) are shown in light grayBack to article page