Deep Artificial Immune Networks for Replay-Free Visual Learning

Researchers developed a Deep Artificial Immune Network using structured, gradient-free immune affinity to enable class-incremental learning without replay. The method utilizes zero-normalized cross-correlation and feature-map binding to maintain memory of earlier classes as new data arrives.