Department of Biomedical & Robotics Engineering
Incheon National University
Computational Science Laboratory
Our Computational Science Laboratory is currently accepting applications for Ph.D., MS, and Undergraduate research positions. These opportunities involve conducting cutting-edge research in mathematical optimization and machine/deep learning to address complex scientific and engineering challenges.
Coming Soon
Computational Science Laboratory
Our research focuses on developing innovative computational methods at the intersection of machine learning, optimization, and scientific discovery. We aim to create intelligent systems that can autonomously make optimal decisions in complex scientific domains.
Our core research focuses on the principled representation of uncertainty in deep learning models. By treating model parameters as probability distributions, we aim to develop models that are not only accurate but also aware of their own confidence. This is crucial for applications where reliability and trust are paramount.
Development of novel deep learning frameworks to significantly speed up the computationally expensive processes of UQ and OED.
Exploration and development of sophisticated BNN architectures and inference techniques to capture both aleatoric (data) and epistemic (model) uncertainty.
Fine-tuning Large Language Models (LLMs) using innovative Bayesian techniques, such as Bayesian LoRA, to enhance performance while robustly quantifying predictive uncertainty.
Investigating and applying methods like semi-supervised learning to improve the performance of OED in scenarios with limited labeled data.
We aim to create intelligent systems that can autonomously make optimal decisions in complex, high-dimensional spaces. This involves the integration of deep learning, reinforcement learning, and classical optimization methods to guide scientific campaigns.
Development of deep learning surrogate models to accelerate the identification of promising candidates within massive molecular libraries. Replacing computationally intensive quantum calculations (e.g., DFT) with highly accurate, rapid predictive models to maximize screening throughput and minimize computational resource allocation.
Design of highly efficient molecular screening strategies by integrating mathematical optimization and Bayesian frameworks. Utilization of Multi-objective Bayesian Optimization (MOBO) to navigate competing properties, such as binding affinity and toxicity, automating the decision-making process to identify superior candidates under budgetary constraints.
Application of state-of-the-art generative models, including Transformers and Large Language Models (LLMs), for the ground-up design of novel molecules. Focused development of innovative molecular structures optimized for specific applications, such as carbon capture, by efficiently navigating the chemical space to find compounds with desired properties.
Development of reinforcement learning agents capable of learning optimal screening policies within HTVS pipelines, maximizing the efficiency of the discovery process.
We apply our expertise in deep learning to solve critical problems across various scientific and engineering disciplines, with a focus on creating practical and impactful solutions.
Fusion of information from diverse medical imaging modalities using advanced deep learning architectures to improve diagnostic accuracy and automated analysis.
Addressing a wide array of problems in science and engineering by custom-designing and applying appropriate deep learning models, contributing to advancements in multiple fields.
Incheon National University (Internal Research Project)
Incheon National University (Internal Research Project)