Research map / Probabilistic and scientific modeling / Dynamical systems
Gamma-Models: Generative Temporal Difference Learning for Infinite-Horizon Prediction
Michael Janner, Igor Mordatch, Sergey Levine
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 training discount gamma from 0 to 0.95 (ablation) | flow gamma-model · Acrobot and pendulum prediction | predictive lookahead (qualitative) | Improved |
| use normalizing flow instead of GAN parameterization (ablation) | gamma-model · Acrobot and pendulum prediction | prediction reliability (qualitative) | Improved |
| use gamma-model for value estimation instead of single-step model rollouts | value estimation · Pendulum, gamma=0.99 | rollout steps: 299-step rollouts -> single feedforward pass | Improved |
| replace SAC value estimation with gamma-MVE | SAC · Continuous control benchmarks | sample efficiency | Improved |
| original GAN vs least-squares GAN objective (ablation) | GAN gamma-model · gamma-model training | training effectiveness | No clear effect |
Closest papers by meaning
- Gamma-Nets: Generalizing Value Estimation over Timescale (AAAI 2020 · similarity 0.88)
- Temporal Difference Flows (ICML 2025 · similarity 0.86)
- Adaptive Quantization in Generative Flow Networks for Probabilistic Sequential Prediction (NeurIPS 2025 · similarity 0.85)
- Learning to Be Uncertain: Pre-training World Models with Horizon-Calibrated Uncertainty (ICLR 2026 · similarity 0.85)
- Self-Improving World Modelling with Latent Actions (NeurIPS 2026 · similarity 0.84)
- Zero-shot World Models via Search in Memory (NeurIPS 2025 · similarity 0.83)
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.
