A simulation-based inference technique that trains neural networks to estimate likelihood ratios for posterior inference without explicit likelihood computation.