QG-MIL Gated Transformer Stabilizes Attention in Medical Imaging MIL

QG-MIL combines RMSNorm pre-normalization, per-head QK normalization, attention output gating, and SwiGLU feed-forward layers to reduce attention concentration in Multiple Instance Learning aggregators. Evaluated across six benchmarks covering whole-slide pathology and hematology, top variants outperform leading baselines without auxiliary losses or multi-stage regularization.