Namjoon Suh

Background

About Namjoon Suh

I am actively seeking a research scientist position in California related to data science, machine learning, or statistics. Followings are my research agenda these days: Tabular Data / Diffusion model / Time Series / Privacy / Mult-modal Feb 2023 – Current Ongoing Research Topics (Privacy, Tabular Foundation Model, Multi-modal Tabular data): o Build a reliable time series tabular synthesizer with privacy guarantees. o Build a tabular synthesizer that can learn the representations of multiple tables via transfer learning. o Work on tabular-image representation learning for the imputation of missing data. Research Topic: AutoDiff: diffusion-based tabular data synthesizer with i.i.d. rows o Lead building a new diffusion-based tabular synthesizer, combining auto-encoder and diffusion model. o Capture correlations of heterogeneous features in table; the main challenge in SOTA tabular synthesizers. o Beat SOTA models under various evaluation metrics; F1, AUROC, Accuracy, RMSE, R-square for 16 dataset. Research Topic: TimeAutoDiff: diffusion-based time series tabular data synthesizer with dependent rows o Build a versatile model for time series tabular data generation across domains finance and healthcare. o Obtain fast sampling time with high, fidelity and utility for single/multi-sequence timeseries tabular data generation, under discriminative, predictive, and feature correlation score metrics. o Enable conditional sampling for counterfactual scenario explorations. For instance, given the sequential data over certain timestamps (such as temperature, weather info) as conditions, the model outputs the values of the heterogeneous variables of interests over the corresponding timestamps (such as traffic volume).