Research map / Probabilistic and scientific modeling / Dynamical systems
Predictive-State Decoders: Encoding the Future into Recurrent 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 |
|---|---|---|---|
| Add Predictive-State Decoders | TRPO · Swimmer | average return: 91.3 \pm 25.5 -> 97.0 \pm 19.4; 6.30% | Helped |
| Add Predictive-State Decoders | TRPO · HalfCheetah | average return: 330 \pm 158 -> 372 \pm 143; 13.0% | Helped |
| Add Predictive-State Decoders | TRPO · Hopper | average return: 1103 \pm 264 -> 1195 \pm 272; 9.06% | Helped |
| Add Predictive-State Decoders | TRPO · Walker2d | average return: 383 \pm 96 -> 416 \pm 88; 8.59% | Helped |
| Add Predictive-State Decoders | TRPO · Walker2d (5e4 steps/iteration) | average return: 1396 \pm 396 -> 1611 \pm 436; 15.4% | Helped |
| Add Predictive-State Decoders | AggreVaTeD LSTM · CartPole | cumulative reward: 73% relative improvement | Helped |
| Add Predictive-State Decoders | AggreVaTeD GRU · CartPole | cumulative reward: 42% relative improvement | Helped |
| Use second-order features instead of identity featurization (ablation) | Predictive-State Decoders · Probabilistic filtering | filtering performance | Hurt |
| Increase predictive-state horizon k (ablation) | AggreVaTeD with Predictive-State Decoders · CartPole and Acrobot | cumulative reward | Helped |
| Add Predictive-State Decoders | TRPO with Basic RNN · InvertedPendulum | average return: 820 \pm 139 -> 820 \pm 118; -0.08% | No clear effect |
| Add Predictive-State Decoders | TRPO with GRU · InvertedPendulum | average return: 673 \pm 268 -> 782 \pm 183; 20.4% | Helped |
| Add Predictive-State Decoders | TRPO with LSTM · InvertedPendulum | average return: 640 \pm 265 -> 784 \pm 215; 22.6% | Helped |
Closest papers by meaning
- Predictive State Recurrent Neural Networks (NeurIPS 2017 · similarity 0.93)
- Recurrent Predictive State Policy Networks (ICML 2018 · similarity 0.90)
- Initialization matters: Orthogonal Predictive State Recurrent Neural Networks (ICLR 2018 · similarity 0.89)
- Mechanistic Interpretability of RNNs emulating Hidden Markov Models (NeurIPS 2025 · similarity 0.87)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.87)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.86)
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.