LRIM benchmark accepted at ICLR 2026

Our paper LRIM: a Physics-Based Benchmark for Provably Evaluating Long-Range Capabilities in Graph Learning has been accepted for poster presentation at ICLR 2026, Rio de Janeiro, Brazil (April 23–27, 2026). We introduce the Long-Range Ising Model Graph Benchmark, where ground truth provably depends on long-range dependencies; the benchmark scales from 256 to 65k nodes and allows controllable long-rangedness. Co-authors: Joël Mathys, Henrik Christiansen, Federico Errica, Takashi Maruyama.