Research map / Architectures and efficiency / RNNs and normalization
Particle Filter Recurrent Neural Networks
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 |
|---|---|---|---|
| increase particles from 1 to 30 (ablation) | PF-LSTM · Avg classification | accuracy (%): 85.88 -> 93.28 | Helped |
| increase particles from 1 to 30 (ablation) | PF-LSTM · Avg regression | scaled regression error: 0.81 -> 0.51 | Helped |
| add soft resampling (ablation) | PF-LSTM · Avg classification | accuracy (%): 89.92 -> 93.28 | Helped |
| add soft resampling (ablation) | PF-LSTM · Avg regression | scaled regression error: 0.76 -> 0.51 | Helped |
| use ReLU with batch normalization instead of tanh (ablation) | PF-LSTM · Avg classification | accuracy (%): 88.55 -> 93.28 | Helped |
| add ELBO loss to prediction loss (ablation) | PF-LSTM · Avg classification | accuracy (%): 90.70 -> 93.28 | Helped |
| add prediction loss to ELBO-only training (ablation) | PF-LSTM · Avg classification | accuracy (%): 88.60 -> 93.28 | Helped |
| add ReLU+BN to plain LSTM (no particles) | LSTM · Avg classification | accuracy (%): 86.28 vs PF-LSTM 93.28 | No clear effect |
| replace LSTM with PF-LSTM (30 particles, similar params) | LSTM · NASDAQ | regression loss: 37.33 -> 4.65 | Helped |
| replace GRU with PF-GRU | GRU · AIR | regression loss: 17.7 -> 19.32 | Hurt |
| replace GRU with PF-GRU | GRU · GAS | accuracy (%): 76.8 -> 83.3 | Helped |
| replace LSTM with PF-LSTM | LSTM · R52 | accuracy (%): 81.1 -> 89.1 | Helped |
Closest papers by meaning
- Predictive-State Decoders: Encoding the Future into Recurrent Networks (NeurIPS 2017 · similarity 0.89)
- Predictive State Recurrent Neural Networks (NeurIPS 2017 · similarity 0.88)
- Recurrent Kalman Networks: Factorized Inference in High-Dimensional Deep Feature Spaces (ICML 2019 · similarity 0.87)
- Initialization matters: Orthogonal Predictive State Recurrent Neural Networks (ICLR 2018 · similarity 0.86)
- Fast-Slow Recurrent Neural Networks (NeurIPS 2017 · similarity 0.86)
- BARNN: A Bayesian Autoregressive and Recurrent Neural Network (ICML 2025 · similarity 0.85)
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.