EVENT DOSSIER
On September 7, 2026, arXiv cs.CV published a paper titled “Synergistic Information Disentanglement for Omni-modal Slide Representation Learning in Computational Pathology”. This study addresses the problem of multi-modal slide representation learning in computational pathology and proposes a method for synergistic information decoupling. The method aims to separate and integrate key information from multi-modal data to optimize the feature representation of pathological slides, thereby improving the performance of downstream diagnostic or analysis tasks.
研究人员提出$\mathrm{\Phi}$-Omni 框架，旨在解决计算病理中多模态自监督学习因强制对齐导致的模态坍塌问题。该框架基于部分信息分解（PID）理论，利用提出的$\mathrm{\Phi}\text{ID}$目标函数抑制边际冗余并最大化不可约协同性，从而蒸馏高阶跨模态交互信号。模型在乳腺（$n$=1031）和肺（$n$=919）队列上进行预训练后，在五个独立外部数据集涵盖的八项任务中，相比监督学习和自监督基线均展现出更优的少样本性能。