<p>High-entropy alloys (HEAs), as an emerging material system that subvert the traditional alloy design concept, have triggered a paradigm shift in the field of materials science with their multi-principal element synergy effect and excellent properties. However, its challenges such as the complexity of atomic-scale interactions, the difficulty of multi-scale phase stability prediction, the cross-scale correlation mechanism of mechanical properties, and the nearly infinite reverse design of the composition space have restricted its systematic development. Machine learning (ML) has significantly improved the simulation accuracy and design efficiency of high-entropy alloys by constructing “atomic-mesoscopic-macroscopic” cross-scale models: Atomic potential functions based on symmetry constraints (such as SNAP and GAP) have achieved molecular dynamics simulations with near-density functional theory accuracy, revealing the strengthening mechanisms of chemical short programs (SRO) and nano-precipitated phases; Random forests are quantitatively correlated with models such as graph neural networks in terms of component-structure-performance, guiding the collaborative optimization of strength and toughness. The reverse design framework driven by Generative Adversarial Network (GAN) and variational autoencoder (VAE) shortens the alloy development cycle by 50%. In the future, the development of multimodal data fusion, physically constrained machine learning models and dedicated software toolchains will accelerate the engineering applications of high-entropy alloys in aerospace, energy catalysis and other fields. </p>