Machine Learning for Science
Machine learning has resulted in several advances for several industry use cases and offers potential to accelerate science simulations and new scientific discoveries.
My specific interests lie in developing machine learning based surrogate models for scientific simulations, especially on problems that require very large amount of data and/or compute. The science data sets the I work with involves heterogeneous systems with different materials, and physical properties such as temeperature, and density. I am also interested in accelerating machine learning approaches on novel hardware architectures including but not limited to GPUs.
- A Comparison of Spectral and Spatial Graph Convolutional Neural Network Kernels Using GraphSAGE-Sparse
- Computational Challenges in the development of a surrogate model for Density Functional Theory
- Machine Learning for Materials Modeling
- Scientific Machine Learning and Data flow acceleration: ASCR HQ Update
- Training-free hyperparameter optimization of neural networks for electronic structures in matter