Research
Research
My research focuses on developing advanced probabilistic models for analyzing spatio-temporal data, with a particular emphasis on extending state-space models to address real-world challenges. Specifically, I am interested in:
Switching Linear Dynamical Systems (SLDS): I work on improving SLDS by incorporating explicit duration modeling, including models with infinite duration support, to enhance segmentation quality and overall model performance. These extensions allow for more flexible and interpretable temporal data analysis.
Bayesian Nonparametric Methods: I leverage Bayesian nonparametrics, such as Hierarchical Dirichlet Process Hidden Markov Models (HDP-HMM), and their advanced extensions to model complex temporal dynamics without requiring a predefined number of hidden states.
Recurrent Models and Explicit Duration Variables: I develop models like Recurrent Explicit Duration Switching Linear Dynamical Systems (REDSLDS), which combine recurrent mechanisms and explicit duration variables to achieve better segmentation and effectively capture complex dynamics.
Efficient Inference Techniques: My work emphasizes advanced inference methods, including Markov Chain Monte Carlo (MCMC) and Pólya-Gamma augmentation, to enable efficient and scalable learning in high-dimensional probabilistic models.
Applications in Spatio-Temporal Phenomena: I apply these models to a range of domains, such as neural activity analysis, vehicle trajectory prediction, and other tasks requiring accurate recognition of temporal patterns.
Through my research, I aim to advance the theoretical foundations of probabilistic modeling while addressing practical challenges in temporal data analysis.
2024
Mikolaj Slupinski, Piotr Lipinski: The Recurrent Sticky Hierarchical Dirichlet Process Hidden Markov Model. CoRR abs/2411.04278 (2024)
Mikolaj Slupinski, Piotr Lipinski: Bayesian Inference in Recurrent Explicit Duration Switching Linear Dynamical Systems. CoRR abs/2411.04280 (2024)
2023
Mikolaj Slupinski, Piotr Lipinski: Improving SLDS Performance Using Explicit Duration Variables with Infinite Support. ICONIP (9) 2023: 112-123
2025
Wenjie Hui: Optimizing Privacy and Efficiency in Autonomous Vehicle Data: Quantization and Knowledge Distillation for Enhanced Anonymization. CoRR abs/2411.04278 (2024)
Natalia Zychowicz: Uncertainty Quantification in Spatiotemporal Geochemical Data Modeling. CoRR abs/2411.04280 (2024)