The Rising Cost of ML Research: Why Yesterday's Breakthroughs Face Rejection Today

Submission volumes to top ML conferences have exploded 7-9 times over the past decade while acceptance rates have flatlined, fundamentally reshaping what it takes to publish. The field has simultaneously codified four new technical demands—ablation studies, rigorous baselines, variance reporting, and reproducibility—that papers from the deep learning era would struggle to meet under contemporary review. This escalating rigor is not merely academic gatekeeping; it directly mirrors the standards production AI systems must now meet.