To make LLMs and VLMs more efficient via compression and
pruning.
DL for Simulation
To accelerate the process of simluation-based science and
engineering via DL.
Past Work
Researched Topics
DL for Drug Discovery
Sep 2022 - Jun 2024
Representation learning to handle various scale of molecules
universally.
Ligand generation using diffusion models for protein target
pockets.
Multimodal model for predicting protein-ligand binding affinity.
During M.S.
Sep 2020 - Aug 2022
Researched Topics
Neural Tangent Kernels (NTKs) and DL theories
NTK arises in the infinite-width limit of neural networks,
simplifying DL models to a linearized form under certain
conditions.
The trend toward larger NNs has driven research on
initialization and training in wide networks, leading to the
infinite-width assumption for analyzing their dynamics.
Modern overparameterized DL models can be analyzed via NTK in
the infinite-width limit as feature dimensions increase.
Generative Models (Diffusion and Schrödinger Bridge
models)
Diffusion models have demonstrated state-of-the-art performance
across various domains.
Schrödinger Bridge (SB) models, based on entropy-regularized
optimal transport, generalize diffusion models.
During B.S.
Mar 2016 - Aug 2020
Undergraduate Research Internship
Topic : Theoretical Background on the Convergence of Optimizers
Used in Deep Learning Mar 2020 - Aug 2020
Topic : Exploring the Mathematical Background Related to Machine
Learning(Reinforcement Learning) Jun 2019 - Aug 2019