About this Abstract |
| Meeting |
2022 TMS Annual Meeting & Exhibition
|
| Symposium
|
Materials Design and Processing Optimization for Advanced Manufacturing: From Fundamentals to Application
|
| Presentation Title |
How Deep Learning Can Help with Materials Design |
| Author(s) |
Sara Kadkhodaei, Ali Davariashtiyani |
| On-Site Speaker (Planned) |
Sara Kadkhodaei |
| Abstract Scope |
In recent years, deep learning has emerged as a successful method for a broad range of artificial intelligence tasks in materials design, especially for learning and prediction of complex structure-property relationships at a variety of scales. While the main utility of deep learning is to predict the structure-property relationships on never-seen-before data, the complex and highly non-linear design of neural networks is prone to overfitting to the training data. Here, I will discuss a number of techniques which we have utilized to make our neural network models generalize better to unseen data. I will present two different models: First is a neural network model that is trained to predict the synthesizability of novel crystalline materials. Second is a neural network model which predicts the formation energy of inorganic crystal compounds. I will discuss the utility of these models in accelerating materials discovery and optimization. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, ICME, Machine Learning |