Permanent magnets are attracting renewed interest in recent years due to increasing demand for strong magnets used in drive motors for hybrid and electric vehicles, wind-power generators etc. It has been more than 30 years since the currently strongest neodymium magnet was invented by Masato Sagawa in 1982. The development of magnet compounds with magnetic properties (saturation magnetization, magnetocrystalline anisotropy, and Curie temperature) exceeding those of Nd₂Fe₁₄B, the main phase of neodymium magnet, is anticipated. The material properties of permanent magnets are determined not only by the intrinsic properties of the main phase; they also depend on the microstructure and grain boundary conditions. Our group combines machine-learning techniques with computational science approaches to develop and apply new methods for exploring magnet compounds. We also analyze data on microstructure and magnetization reversal obtained from experiments and simulations and clarify the coercivity mechanisms.