Abstract
Two-dimensional materials with flat electronic bands are promising for realising exotic quantum phenomena such as unconventional superconductivity and nontrivial topology. However, exploring their vast chemical space is a significant challenge. Here we introduce elf, an unsupervised convolutional autoencoder that encodes electronic band structure images into fingerprint vectors, enabling the autonomous clustering of materials by electronic properties beyond traditional chemical paradigms. Unsupervised visualisation of the fingerprint space then uncovers hidden chemical trends and identifies promising candidates based on similarities to well-studied exemplars. This approach complements high-throughput ab initio methods by rapidly screening candidates and guiding further investigations into the mechanisms underlying flat-band physics. The elf autoencoder is a powerful tool for autonomous discovery of unexplored flat-band materials, enabling unbiased identification of compounds with desirable electronic properties across the 2D chemical space.