Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Entropic Desired Dynamics for Intrinsic Control
Steven Stenberg Hansen, Guillaume Desjardins, Kate Baumli, David Warde-Farley, Nicolas Heess, Simon Osindero, Volodymyr Mnih
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 |
|---|---|---|---|
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Montezuma's Revenge | average return: 0 -> 30.9 | Improved |
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Montezuma's Revenge | lifetime coverage (x10^3): 4.54 -> 8.07 | Improved |
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Berzerk | lifetime coverage (x10^3): 10.6 -> 61.3 | Improved |
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Berzerk | episodic coverage and average return: 0.562 -> 0.382 episodic; 584 -> 477 return | Hurt |
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Hero | average return: 68.7 -> 1.34k | Improved |
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Private Eye | average return: -43.6 -> 1.07k | Improved |
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Ms. Pacman | average return: 360 -> 587 | Improved |
| add additive latent dynamics z_i+Delta_i (EDDICT) in place of independent per-option codes (EDDICT-∆) (ablation) | EDDICT · Seaquest | average return: 238 -> 400 | Improved |
| use globally consistent Markov-chain codes (EDDICT) instead of relative-only codes of RVIC | EDDICT · Montezuma's Revenge | average return: 0 -> 30.9 | Improved |
| use EDDICT instead of RVIC | EDDICT · Berzerk | lifetime coverage (x10^3): 31.3 -> 61.3 | Improved |
| use EDDICT instead of RVIC | EDDICT · Hero | episodic coverage and average return: 2.58 -> 2.1 episodic; 805 -> 1.34k return | Mixed |
| use EDDICT instead of RVIC | EDDICT · Private Eye | episodic coverage and average return: 5.41 -> 4.86 episodic; 886 -> 1.07k return | Mixed |
Closest papers by meaning
- Simplifying Latent Dynamics with Softly State-Invariant World Models (NeurIPS 2024 · similarity 0.87)
- ReLaX: Reasoning with Latent Exploration for Large Reasoning Models (CVPR 2026 · similarity 0.85)
- Reinforcement Learning Under Latent Dynamics: Toward Statistical and Algorithmic Modularity (NeurIPS 2024 · similarity 0.85)
- Action-Dependent Optimality-Preserving Reward Shaping (ICML 2025 · similarity 0.84)
- Multi-Agent Learning under Uncertainty: Recurrence vs. Concentration (NeurIPS 2025 · similarity 0.84)
- PcLast: Discovering Plannable Continuous Latent States (ICML 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.
