Research map / Probabilistic and scientific modeling / Variational and generative sampling
Backprop KF: Learning Discriminative Deterministic State Estimators
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| End-to-end training of the Kalman filter with CNN observation and learned observation covariance (BKF) vs. piecewise-trained KF | Kalman filter with CNN observation model · Synthetic tracking | RMS test error: 0.1160 -> 0.0537 | Helped |
| Replacing the Kalman filter with a generic LSTM (vs BKF) | recurrent state estimator · Synthetic tracking | RMS test error: 0.0537 -> 0.1407 (64 units) | Hurt |
| Incorporating a Kalman filter structure vs. feedforward-only estimation | feedforward CNN estimator · Synthetic tracking | RMS test error: 0.2322 -> 0.0537 | Helped |
| End-to-end BKF vs piecewise KF on KITTI visual odometry, 10 training trajectories | Kalman filter with CNN observation model · KITTI Test 100, 10 trajectories | Translational error [m/m]: 0.2265 -> 0.2062 | Helped |
| End-to-end BKF vs piecewise KF on KITTI visual odometry, 3 training trajectories | Kalman filter with CNN observation model · KITTI Test 100, 3 trajectories | Translational error [m/m]: 0.3257 -> 0.3089 | Helped |
| BKF vs LSTM on KITTI visual odometry (small data) | recurrent state estimator · KITTI Test 100, 3 trajectories | Translational error [m/m]: 0.5199 -> 0.3089 (vs LSTM 256 units) | Helped |
| End-to-end BKF vs piecewise KF on KITTI rotational error, 10 trajectories | Kalman filter with CNN observation model · KITTI Test 100, 10 trajectories | Rotational error [deg/m]: 0.0978 -> 0.0801 | Helped |
| Learning observation covariance conditioned on image (BKF) vs fixed covariance (piecewise) | Kalman filter with CNN observation model · Synthetic tracking (difficulty sweep) | tracking error | No clear effect |
Closest papers by meaning
- Recurrent Kalman Networks: Factorized Inference in High-Dimensional Deep Feature Spaces (ICML 2019 · similarity 0.88)
- Deep Tracking: Seeing Beyond Seeing Using Recurrent Neural Networks (AAAI 2016 · similarity 0.87)
- Predictively-Oriented Kalman Filtering (NeurIPS 2026 · similarity 0.85)
- Learning to be Smooth: An End-to-End Differentiable Particle Smoother (NeurIPS 2024 · similarity 0.85)
- Gated Inference Network: Inference and Learning State-Space Models (NeurIPS 2024 · similarity 0.84)
- From Variance to Veracity: Unbundling and Mitigating Gradient Variance in Differentiable Bundle Adjustment Layers (CVPR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.