A Variational Autoencoder for Neural Temporal Point Processes with Dynamic Latent Graphs
This addresses the challenge of accurately predicting event sequences with time-varying dependencies, which is incremental as it builds on existing neural point process methods by incorporating dynamic latent graphs.
The paper tackles the problem of modeling continuously-observed event sequences with dynamic temporal dependencies, proposing a variational autoencoder that partitions time into sub-intervals to capture changing event dynamics and dependency graphs, resulting in higher accuracy in predicting inter-event times and event types compared to existing state-of-the-art neural point processes.
Continuously-observed event occurrences, often exhibit self- and mutually-exciting effects, which can be well modeled using temporal point processes. Beyond that, these event dynamics may also change over time, with certain periodic trends. We propose a novel variational auto-encoder to capture such a mixture of temporal dynamics. More specifically, the whole time interval of the input sequence is partitioned into a set of sub-intervals. The event dynamics are assumed to be stationary within each sub-interval, but could be changing across those sub-intervals. In particular, we use a sequential latent variable model to learn a dependency graph between the observed dimensions, for each sub-interval. The model predicts the future event times, by using the learned dependency graph to remove the noncontributing influences of past events. By doing so, the proposed model demonstrates its higher accuracy in predicting inter-event times and event types for several real-world event sequences, compared with existing state of the art neural point processes.