Poster
in
Workshop: AI for Science: Scaling in AI for Scientific Discovery
Multi-task Extension of Geometrically Aligned Transfer Encoder
Sung Moon Ko · Sumin Lee · Dae-Woong Jeong · Hyunseung Kim · Chanhui Lee · Soorin Yim · Sehui Han
Keywords: [ Regression ] [ Molecular Property Prediction ] [ Multi-task Learning ] [ geometrical deep learning ]
Molecular datasets often suffer from a lack of data. It is well-known that gathering data is difficult due to the complexity of experimentation or simulation involved. Here, we leverage mutual information across different tasks in molecular data to address this issue. We extend an algorithm that utilizes the geometric characteristics of the encoding space, known as the Geometrically Aligned Transfer Encoder (GATE), to a multi-task setup. Thus, we connect multiple molecular tasks by aligning the curved coordinates onto locally flat coordinates, ensuring the flow of information from source tasks to target data to support the performance.