Hi, would you mind to share the code to get the figure1.
As described in the paper, 'To illustrate this issue, we consider three classes from CIFAR10: dog, cat, and plane. We train a KCL model [23] on this data for different imbalance ratios, ρ. For visualization clarity we use a 2D feature space. As seen in Fig. 1(a), when the classes are balanced (i.e., ρ=1:1:1), the centers of the three classes are uniformly distributed in the KCL feature space. In contrast, when the imbalance ratio is high (e.g., ρ=100:1:1), the classes with fewer training instances start to collapse into each other, leading to unclear and inseparable decision boundaries, and thus lower performance.'?
Hi, would you mind to share the code to get the figure1. As described in the paper, 'To illustrate this issue, we consider three classes from CIFAR10: dog, cat, and plane. We train a KCL model [23] on this data for different imbalance ratios, ρ. For visualization clarity we use a 2D feature space. As seen in Fig. 1(a), when the classes are balanced (i.e., ρ=1:1:1), the centers of the three classes are uniformly distributed in the KCL feature space. In contrast, when the imbalance ratio is high (e.g., ρ=100:1:1), the classes with fewer training instances start to collapse into each other, leading to unclear and inseparable decision boundaries, and thus lower performance.'?