resume

Pursuing a bachelor's degree in Computer Science and Engineering.

Increasing Gmail security by implementing an attachment checker for client-side encrypted Gmail accounts, to block potentially harmful file types from being attached or downloaded.

Improved Google search results by creating large language machine learning (ML) models for detecting badly formed queries and developing syntactic filters in C++.

Designing novel materials by developing a Euclidean symmetry (E(3))-equivariant autoregressive generative ML model for constructing 3D molecular structures, using the e3nn-jax Python library. Also helped develop/update an E(3)-equivariant machine learning model to characterize the magnetic properties of crystalline materials.

Prepared, trained, tuned an XGBoost ML model in Python (sklearn) to predict company attritions; analyzed internal financial/demographic/customer/employee satisfaction data using Python, SQL, and Excel.

Created a computational simulation of vesicular stomatitis virus ribonucleoprotein motion within host cells, using Python.

Symphony: Symmetry-Equivariant Point-Centered Spherical Harmonics for Molecule Generation

Daigavane A, Kim S, Geiger M, Smidt T

12th International Conference on Learning Representations, 11/2023

We present Symphony, an E(3)-equivariant autoregressive generative model for 3D molecular geometries that iteratively builds a molecule from molecular fragments. Existing autoregressive models such as G-SchNet and G-SphereNet for molecules utilize rotationally invariant features to respect the 3D symmetries of molecules. In contrast, Symphony uses message-passing with higher-degree E(3)-equivariant features. This allows a novel representation of probability distributions via spherical harmonic signals to efficiently model the 3D geometry of molecules. We show that Symphony is able to accurately generate small molecules from the QM9 dataset, outperforming existing autoregressive models and approaching the performance of diffusion models.

Machine learning magnetism classifiers from atomic coordinates

Merker HA, Heiberger H, Nguyen L, Liu T, Chen Z, Andrejevic N, Drucker NC, Okabe R, Kim S, Wang Y, Smidt T, Li M.

iScience, 11/2022

The determination of magnetic structure poses a long-standing challenge in condensed matter physics and materials science. Experimental techniques such as neutron diffraction are resource-limited and require complex structure refinement protocols, while computational approaches such as first-principles density functional theory (DFT) need additional semi-empirical correction, and reliable prediction is still largely limited to collinear magnetism. Here, we present a machine learning model that aims to classify the magnetic structure by inputting atomic coordinates containing transition metal and rare earth elements. By building a Euclidean equivariant neural network that preserves the crystallographic symmetry, the magnetic structure (ferromagnetic, antiferromagnetic, and non-magnetic) and magnetic propagation vector (zero or non-zero) can be predicted with an average accuracy of 77.8% and 73.6%. In particular, a 91% accuracy is reached when predicting no magnetic ordering even if the structure contains magnetic element(s). Our work represents one step forward to solving the grand challenge of full magnetic structure determination.

Managed the development of a new Django website for ESP, including the framework needed to run middle-/high-school educational programs. Wrote documentation for the current website’s code base.

Directed and organized a 5-week educational program for middle and high schoolers in the Boston area, with 25 classes and 460 students.