Summary and Contributions: This work first proposes a new parametrization for several classic temporal point processes (TPPs), which enables efficient parallel likelihood computation and sampling. TPP allows to naturally handle data that consists of variable-number events in continuous time. These classic TTP models with existing parametrization was inherently sequential. Next, the authors proposed a new class of non-recurrent TPP models, namely TriTPP, where both sampling and likelihood computation can be done in parallel. TPP models combined with recurrent neural networks provide a highly flexible powerful framework, but still remain sequential, making TPPs poorly suited for sampling.

Temporal point process (TPP) models combined with recurrent neural networks provide a powerful framework for modeling continuous-time event data. While such models are flexible, they are inherently sequential and therefore cannot benefit from the parallelism of modern hardware. By exploiting the recent developments in the field of normalizing flows, we design TriTPP - a new class of non-recurrent TPP models, where both sampling and likelihood computation can be done in parallel. TriTPP matches the flexibility of RNN-based methods but permits several orders of magnitude faster sampling. This enables us to use the new model for variational inference in continuous-time discrete-state systems.