Michael Wieck-Sosa

I am a fourth-year PhD student in Statistics at Carnegie Mellon University.

I am very lucky to be advised by Aaditya Ramdas and Cosma Shalizi.

Here is my resume. You can reach me at mwiecksosa AT cmu DOT edu.

My research interests are at the intersection of:

• Causality (invariant causal prediction, conditional independence testing, causal structure learning, mechanism discovery)
• Dependence (high-dimensional time series, time-varying networks, spatiotemporal data, nonlinear stochastic dynamics)
• Machine learning (deep neural networks, transformers, foundation models, large language models)

Research

Papers

• Estimating dynamic models by matching random features (with Cosma Shalizi). Submitted to the Journal of the American Statistical Association. arXiv:2607.21916. Code.
• Dynamic models with p parameters are identified by 2p+1 random features (with Cosma Shalizi). Submitted to Physical Review E. arXiv:2607.16035. Code.
• The dynamic generalized covariance measure for conditional independence testing with nonstationary time series (with Michel F. C. Haddad and Aaditya Ramdas). Minor revision at the Journal of Business & Economic Statistics. arXiv:2504.21647. Slides. Code.

Current projects

• Sequential invariant causal prediction for time series (with Aaditya Ramdas).
• Learning generative sequence models with infinite memory (with Cosma Shalizi).
• Learning the signature of a large language model (with Cosma Shalizi).
• Simulation-based inference through random features (with Cosma Shalizi).
• How foundation models learn temporal dependencies in practice (with Tom Zhang* and Chad Schafer).

* Denotes students advised.

News

• June 2026: I gave an invited talk at the 9th International Workshop in Sequential Methodologies (IWSM 2026).
• April 2026: I was awarded the DeGroot-Goel Fellowship for 2026 by the CMU Statistics & Data Science Department faculty.
• April 2026: I successfully proposed my thesis. Topics: generative modeling and theory/methods for dependent data.