Researchers have developed a new approach to artificial immune networks (AINs) that can learn from images without needing labels or backpropagation. The method, called Deep AIN, uses structured affinity and preserves spatial structure in visual data. This allows the network to adaptively reorganize its latent coordinates as it encounters new classes of images. Experiments on several datasets show that this approach can achieve high accuracy without requiring replay or label-driven updates. --- Why it matters: This matters because it provides a new way for AINs to learn from visual data, potentially leading to more efficient and effective image classification systems. The ability to adaptively reorganize latent coordinates could also improve the network's ability to generalize to new classes of images. Source: https://arxiv.org/abs/2608.20104