Reproduce “Conditional Generative Modeling for High-dimensional Marked Temporal Point Processes” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zheng Dong, Zekai Fan, Shixiang Zhu
This research introduces a Conditional Event Generator (CEG) framework, instantiated with a Conditional Denoising Diffusion Model (CDDM), for modeling high-dimensional marked temporal point processes. The approach bypasses computationally intractable intensity function integrals and inefficient thinning algorithms, enabling significantly faster and more accurate generation of sequential events with complex, high-dimensional associated data.
Key insights
- Directly modeling and generating events conditioned on historical context using modern generative models can overcome the computational and representational challenges of intensity-based point processes for high-dimensional marks.
- A model-agnostic generative framework allows flexible integration of various deep learning generative models, enabling adaptation to complex data and leveraging advancements in generative AI.
- Bypassing explicit intensity function estimation and the thinning algorithm leads to substantial improvements in computational efficiency for both training and event generation, making high-dimensional marked point process modeling practical.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.