Decision-Focused Active Learning Optimizes Critical-Materials Recovery

Active learning policies for selective precipitation of rare-earth elements reached enrichment maximums in 16 to 24 experiments, significantly faster than non-adaptive methods. The study uses a two-stage reconstruction approach to optimize the recovery of neodymium-iron-boron and samarium-cobalt magnets.