Researchers have introduced Latent Block-Diffusion Temporal Point Processes (LBDTPP), a new framework designed for generating asynchronous event sequences. This semi-autoregressive approach combines the benefits of autoregressive models for variable-length output with the parallel generation capabilities of diffusion models. LBDTPP operates by defining an autoregressive distribution over event blocks in latent space and then applying Gaussian diffusion within each block, aiming to reduce error accumulation compared to traditional event-wise autoregressive methods. Experiments on six real-world datasets show LBDTPP outperforming existing Temporal Point Process baselines in both unconditional and conditional generation tasks. AI
IMPACT This framework could improve the quality and efficiency of generating complex, asynchronous event data across various domains.
RANK_REASON The cluster contains an academic paper detailing a new model/framework for event sequence generation.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
