Researchers have developed a machine learning framework to design pixelated millimeter-wave patch antennas. The approach uses a binary classifier to filter out non-resonant patterns before simulating them, and then trains a forward surrogate model to predict the antenna's response across a range of frequencies. This allows for the automatic design and reconfiguration of antenna structures, with results showing good agreement between predicted and simulated responses. --- Why it matters: This work matters because it enables the efficient design of millimeter-wave antennas, which are crucial for high-speed wireless communication systems. The ability to automatically design and optimize these antennas could lead to significant improvements in data transfer rates and network capacity. Source: https://arxiv.org/abs/2608.23469