- Abstract: Problem-specific algorithms and generic machine learning approaches have complementary strengths and weaknesses, trading-off data efficiency and generality. To find the right balance between these, we propose to use problem-specific information encoded in algorithms together with the ability to learn details about the problem-instance from data. We demonstrate this approach in the context of state estimation in robotics, where we propose end-to-end learnable histogram filters---a differentiable implementation of histogram filters that encodes the structure of recursive state estimation using prediction and measurement update but allows the specific models to be learned end-to-end, i.e. in such a way that they optimize the performance of the filter, using either supervised or unsupervised learning.
- TL;DR: a way to combine the algorithmic structure of Bayes filters with the end-to-end learnability of neural networks
- Conflicts: tu-berlin.de
- Keywords: Deep learning, Unsupervised Learning