Curtis McDonald

Curtis McDonald

Postdoctoral Researcher at Simons Institute, UC Berkeley

Publications

Research Areas

MCMC For Machine Learning

I am interested in Markov Chain Monte Carlo (MCMC), optimal transport, and generative models- various methods to produce data from a given distribution. Ultimately, I would like to apply a stochastic approach to train neural networks not using back propagation but treating the parameters as a sampling problem. This converts a non-convex optimization problem into a non-log-concave sampling problem, which has more freedom to explore the state space and not get stuck in local minima.

Key quantities to understand in this context are the computational complexity of sampling algorithms, and bounds on the predictive ability of posterior estimators. Designing and understanding effective MCMC methods for complex machine learning models can solve the computational limitations that hold back the practical use of Bayesian methods, and also shed light on the performance of closely related random optimization algorithms like stochastic gradient descent.

Accuracy of Diffusion Models

Generative models are increasingly applied in agentic AI systems which can have impacts on real world processes outside themselves, but we have limited methods to quantify how far their samples stray from their target distribution. This can lead to mode collapse, reward hacking, and other misaligned behaviour. My research into diffusion models aims to improve their sampling accuracy with better score estimators and replacing intractable terms with closed form expressions derived from symmetries of the diffusion process where possible. Ultimately, I would like to carry the guarantees of MCMC mixing theory into the setting where diffusion models succeed in practice, and establish the statistical-computational tradeoffs of these modern methods.

Stochastic Control

My work in stochastic control focuses on filter stability and robust control for hidden Markov models. Namely, given “bad” initial information, can a decision maker correct this mistake with the arrival of new information and make good control decisions after enough observation of the system.

Applied Data Science

I have also had the opportunity to collaborate on several applied data science projects with colleagues in different fields ranging from social media governance, natural language processing, and sports medicine.

Publications

* Select Publications

Preprints

  • * McDonald, C., Bartlett, P. Internal Guidance as a Sampling Method for Tilted Diffusion Models. Under Review (2026). Link

  • * McDonald, C., Rojas, K., Tao, M., Bartlett, P. Improving score-based sampling via affine post-processing. Under Review (2026). Link

  • McDonald, C. Technical Note on Relating Scores of Tilted Distributions. April, 2026. Link

Journal Articles

  • * McDonald, C., Barron, A. Rapid Bayesian Computation and Estimation for Neural Networks via Log-Concave Coupling. Mathematical Statistics and Learning, 9(3):199–246, 2026. DOI 10.4171/MSL/59 DOI 10.4171/MSL/59 Link

  • McDonald, C., Yuksel, S. Stochastic Observability and Filter Stability under Several Criteria. IEEE Trans. Automatic Control. vol. 69, no. 5, pp. 2931-2946, May 2024. Link

  • McDonald, C., Yuksel, S. Robustness to incorrect priors and controlled filter stability in partially observed stochastic control. SIAM Journal on Control and Optimization. vol 60, issue 2 . April 2022. Link

  • McDonald, C., Yuksel, S. Exponential filter stability via Dobrushin’s coefficient. Electronic Communications in Probability. vol 25, no. 53. pp. 1-13. August 2020. Link

  • McDonald, C., Alajaji, F., Yuksel, S. Two-Way Gaussian Networks with a Jammer and Decentralized Control. IEEE Transactions on Control of Network Systems. vol. 7, no. 1, pp. 446-457, March 2020. Link

Workshop and Conference Papers

  • Dern, N., Mazioud, S., Heiss, J., Ghosh, A., McDonald, C., Clara, G., Yu, B. (2026) Tracking Training Phases in Compositional Learning with Task-Agnostic Measures. CompLearn 2026 Workshop at the International Conference on Machine Learning (ICML). Link

  • McDonald, C., Barron, A. (2024). Log-Concave Coupling for Sampling Neural Net Posteriors. 2024 IEEE International Symposium on Information Theory (ISIT). Link.

  • McDonald, C., Barron, A. (2022) Proposal of a Score Based Approach to Sampling Using Monte Carlo Estimation of Score and Oracle Access to Target Density. NeurIPS 2022 Conference Workshop on Score Based Methods. Link

  • McDonald, C., Yuksel, S. (2019) Observability and Filter Stability for Partially Observed Markov Processes. 2019 IEEE 58th Conference on Decision and Control (CDC). pp. 1623-1628. Link

  • McDonald, C., Yuksel, S. (2018) Stability of Non-Linear Filters, Observability and Relative Entropy. 2018 56th Annual Allerton Conference on Communication, Control, and Computing (Allerton). pp 110 - 114. Link

  • McDonald, C., Alajaji, F., Yuksel, S. (2018) Two-Way Gaussian Channels with an Intelligent Jammer. 2018 Annual American Control Conference (ACC). pp. 1784-1789. Link

Data Science Applications

  • Sieberer J. M., Park N., Manafzadeh A. R., Desroches S. T., Brennan K., McDonald C., Tommasini S. M., Wiznia D. H., Fulkerson J. P. Visualization of Trochlear Dysplasia Using 3-Dimensional Curvature Analysis in Patients With Patellar Instability Facilitates Understanding and Improves the Reliability of the Entry Point to Trochlea Groove Angle. Arthroscopy, Sports Medicine, and Rehabilitation. September, 2024. Link

  • Sieberer J. M., Park N., Rancu A. L., Desroches S. T., McDonald C., Manafzadeh A. R., Tommasini S. M., Wiznia D. H., Fulkerson J. P. Analyzing Alignment Error in Tibial Tuberosity–Trochlear Groove Distance in Clinical Scans Using 2D and 3D Methods. The American Journal of Sports Medicine. September, 2024. Link

  • Beitler B. G., Sieberer J., Islam W., McDonald C., Yu K., Tommasini S. M., Fulkerson J. P. The Morphologic Patella Entry Point Into The Proximal Trochlea Is More Lateral in Recurrent Dislocators Than Controls as Measured by Entry Point-Trochlear Groove Angle. Arthroscopy: The Journal of Arthroscopic and Related Surgery. May, 2024. Link

  • Kim, J., McDonald, C., Meosky, P., Katsaros, M., Tyler, T. Promoting Online Civility Through Platform Architecture. Journal of Online Trust and Safety. vol 1, issue 4. September 2022. Link

  • Kumar N., Corpus I., Hans M., Harle N., Yang N., McDonald C., Nakamura Sakai S., Janmohamed K., Chen K., Altice F. L., Tang W., Schwartz J. L., Jones-Jang S. M., Saha K., Memon S. A., Bauch C. T., Choudhury M. D., Papakyriakopoulos O., Tucker J. D., Goyal A., Tyagi A., Khoshnood K., Omer S. COVID-19 vaccine perceptions in the initial phases of US vaccine roll-out: an observational study on reddit. BMC Public Health. vol 22, issue 1, March 2022. Link

Dissertations

  • McDonald, Curtis James. Computation and Estimation for Neural Networks via Log-Concave Coupling. Diss. Yale University, Department of Statistics and Data Science, 2025. Link.

  • McDonald, Curtis James. Filter Stability, Observability and Robustness for Partially Observed Stochastic Dynamical Systems. Diss. Queen’s University (Canada), 2019. Link