Neural Temporal Point Processes (NTPPs) are powerful for modeling event sequences, but lack rigorous uncertainty quantification. This paper proposes NEXTPP-SDE, integrating SDEs with NTPPs to provide principled uncertainty-aware predictions. The approach models latent dynamics as diffusion processes, enabling quantification of both aleatoric and epistemic uncertainties.
Key findings
NEXTPP-SDE integrates SDEs with NTPPs for uncertainty-aware event prediction.
Models latent dynamics as diffusion processes to quantify aleatoric and epistemic uncertainties.
Proposes a comprehensive method design including architecture, training objectives, and inference procedures.
Limitations & open questions
The paper is a research proposal and does not yet report experimental results.
The effectiveness of NEXTPP-SDE in real-world high-stakes domains remains to be validated.