38 Anomaly Detection with Variational Autoencoders Table 3.2: Lung nodule dataset after data augmentation. Healthy Cancer Total Training 1722 460 2182 Validation 431 115 546 Testing 539 143 682 Figure 3.5: Displaying 25 slices of 28 ×28 pixels, as a representation of the cube of 28×28×28 voxels used for training our models. be part of the experimental setup, and not of the general method itself. However, which architectures work well to properly learn to estimate the likelihood of the data largely depends on the size and properties of the data. In particular, convolutional neural networks (CNNs) are expected to work well for image data. It is important to give the right amount of capacity to the neural networks. Models with too low capacity may not be able to capture the density distribution well enough.
RkJQdWJsaXNoZXIy MjY0ODMw