Z-ENG Leveraging Machine Learning for Predictive Modeling in Non-IT Domains
2024-2025 tavasz
Nincs megadva
Téma leírása
Many Non-IT fields have been considered a very important platform for research in the area of Artificial Intelligence. These fields offer many complex problems requiring measurements that return a huge amount of data. This data can be processed and used to extract useful information that can be used by AI algorithms in order to facilitate, improve, or even substitute the measurement task needed in these fields with a prediction method that is nondestructive and faster.
Directions open this semester:
Prediction of viscosity using droplets videos.
Dissolution Profile Prediction from spectroscopy data
In addition to the standard ML pipeline, students will apply model distillation techniques to simplify complex pre-trained models while retaining their predictive power. Following this, Explainable AI (XAI) methods such as SHAP or LIME will be used to interpret the distilled models, offering insight into how they generate predictions. This will allow students to assess the balance between model performance, simplicity, and interpretability, making complex models more accessible and understandable.
Tasks to be performed by the student will include:
- Present the Machine Learning algorithms which will be used and the environment on which they will be applied
- Study and analyze the collected data, make the necessary preprocessing and visualization
- Try different approaches for predictions to generate varying results and compare them
- Design the environment and implement the algorithm(s)
- Apply model distillation and XAI techniques to the resulting models
- Verify that the distilled models work as intended, are interpretable, and the prediction results remain accurate
Feltételek
Maximális létszám:
5 fő