Portrait placeholder for Mijung Park. Replace this image with a headshot before publishing if desired.

Mi Jung Park

Assistant Professor · Department of Computer Science · University of British Columbia

Create machine learning tools for a better world.

I lead a machine learning research group working on privacy-preserving machine learning, differentially private synthetic data generation, Bayesian learning, and AI safety in diffusion foundation models.

  • Email: mijungp@cs.ubc.ca
  • Office: ICICS/CS X863, University of British Columbia
  • Research areas: differential privacy, synthetic data, generative models, AI safety, Bayesian learning
Publications People CV summary Google Scholar

Welcome

Our lab develops privacy- and safety-preserving machine learning methods for sensitive, high-dimensional settings, with a probabilistic perspective grounded in Bayesian inference, kernel methods, uncertainty quantification, and generative modelling. We have made foundational contributions to differentially private synthetic data generation, including DP-MERF, DP-HP (Hermite-Polynomials), and DP-LDMs (latent diffusion models). We have also developed privacy-preserving inference and testing methods, including Variational Bayes in Private Settings, DP kernel two-sample testing, and interpretable & DP prediction. Our recent work extends this program to safe diffusion foundation models, including Training-Free Safe Denoisers, Safety-Guided Flow, and Safety-Aware Denoising for text diffusion. We are also interested in applying Bayesian perspectives to modern machine learning systems.

News

Selected research directions

Safe foundation models

Developing methods for safer generation in diffusion and text-diffusion models, including safety-aware denoisers and negative guidance.

Privacy-preserving data generation

Designing differentially private algorithms for synthetic data generation and data distillation.

Privacy, interpretability, and fairness

Studying how privacy-preserving learning changes model behavior, interpretability, fairness, and causal analysis.

Bayesian machine learning

Applying Bayesian methods to efficient inference, neural data analysis, neural network compression, and foundation-model prompting.

Selected publications

  1. The Safety-Aware Denoiser for Text Diffusion Models. Amman Yusuf, Zhejun Jiang, and Mijung Park. ICML 2026.
  2. Safety-Guided Flow: A Unified Framework For Negative Guidance In Safe Generation. Mingyu Kim, Young Heon Kim, and Mijung Park. ICLR 2026, oral.
  3. Training-Free Safe Denoisers for Safe Use of Diffusion Models. Mingyu Kim, Dongjun Kim, Amman Yusuf, Stefano Ermon, and Mijung Park. NeurIPS 2025.
  4. Differentially Private Neural Tangent Kernels for Privacy-Preserving Data Generation. Yilin Yang, Kamil Adamczewski, Xiaoxiao Li, Danica J. Sutherland, and Mijung Park. JAIR 2025.
  5. Differentially Private Latent Diffusion Models. Michael F. Liu, Saiyue Lyu, Margarita Vinaroz, and Mijung Park. TMLR 2024.

See more publications →

Teaching

Teaching details →