The project has been realized in colaboration with two of my colleagues from Universität des Saarlandes during the Winter Term 2024/2025 for the subject Neural Networks: Theory and Implementation.
The scope of the project is to train and further analyze possible techniques to have a model learn and predict the correct value of lipophilicity for any different molecule structure.
The project consists of 3 main parts:
- Fine-tuning Chemical Models. It includes data exploration along with further training starting from a pre-trained model and fine-tuning.
- Influence function-based Data Selection. The purpose of this section is to improve the model performance by including external data.
- Exploration of Data Selection & Fine-Tuning Models. The purpose of this section is to explore alternative methods for data selection and to investigate several fine-tuning techniques to adapt a pre-trained model for a specific task.