Differentiable Particle Filters through Conditional Normalizing Flow
This work addresses a specific bottleneck in particle filtering for researchers in machine learning and signal processing, offering an incremental improvement over existing methods.
The paper tackled the problem of improving differentiable particle filters by using conditional normalizing flows to construct better proposal distributions and dynamic models, resulting in enhanced performance demonstrated in a visual tracking task.
Differentiable particle filters provide a flexible mechanism to adaptively train dynamic and measurement models by learning from observed data. However, most existing differentiable particle filters are within the bootstrap particle filtering framework and fail to incorporate the information from latest observations to construct better proposals. In this paper, we utilize conditional normalizing flows to construct proposal distributions for differentiable particle filters, enriching the distribution families that the proposal distributions can represent. In addition, normalizing flows are incorporated in the construction of the dynamic model, resulting in a more expressive dynamic model. We demonstrate the performance of the proposed conditional normalizing flow-based differentiable particle filters in a visual tracking task.