ICML 2024 | Unified contrastive learning method for multi view molecular representation learning

Huan Welcome Submit original articles of literature interpretation to this official account. Email: 380198025@q q.com Please send the manuscript as an attachment. For articles related to enrollment, study tours, and recruitment both domestically and internationally, please contact WeChat: xiongzhankun1997.

Compile| Ni Yujing

Review | Wang Yongkang

Today, we are sharing an article titled “UniCorn: A Unified Contrastive Learning Approach for Multi view Molecular Representation Learning” published by Professor Lan Yanyan’s research team at the Intelligent Industry Research Institute of Tsinghua University at ICML 2024. This article proposes a unified molecular pre training framework based on contrastive learning.

abstract

Recently, developing pre trained base models in the fields of CV and NLP has become a clear trend. However, for molecular pre training, there is a lack of a universal model that can be effectively applied to various molecular tasks, and a lack of deep understanding of existing pre training methods, including 2D image masking, 2D-3D contrastive learning, and 3D denoising Such methods are not conducive to the development of molecular based models. This article…