Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Contrastive Active Inference
Pietro Mazzaglia, Tim Verbelen, Bart Dhoedt
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 |
|---|---|---|---|
| replace likelihood (reconstruction) representation model with contrastive one | Active inference agent · representation model compute | MMACs: 212.2 -> 15.4 | Improved |
| replace likelihood representation model with contrastive one | Active inference agent · representation model size | # Params: 4485.7k -> 1266.7k | Improved |
| use contrastive instead of reconstruction-based agent | Dreamer / AIF · wall-clock time vs Dreamer | relative time: 1.00 -> 0.84 | Improved |
| use observation reconstruction for AIF instead of contrastive | Active inference agent · wall-clock time vs Dreamer | relative time: 0.84 -> 3.24 | Hurt |
| use contrastive AIF instead of likelihood AIF | Active inference agent · MiniGrid Empty 8x8 | reward consistency | Improved |
| use contrastive AIF instead of likelihood AIF | Active inference agent · MiniGrid Empty 6x6 | mean reward: below 0.4 -> comparable to MBRL | Improved |
| use contrastive AIF instead of Dreamer | Active inference agent · Reacher Easy and Hard | convergence speed | Improved |
| use likelihood AIF instead of contrastive AIF | Active inference agent · Reacher Easy/Hard | reward | Hurt |
| use Dreamer (reconstruction) instead of contrastive models | MBRL · Distracting Reacher Easy | reward | Hurt |
| use contrastive Dreamer on distracting backgrounds | Dreamer · Distracting Reacher Easy | reward | Improved |
| use Contrastive-AIF with unchanged goal image under background change | Active inference agent · Distracting Reacher Easy | reward | Improved |
| use contrastive utility instead of likelihood utility | Active inference agent · Reacher Hard utility values | utility discrimination: 0.98 vs 1.00 | Improved |
Closest papers by meaning
- Deep active inference agents using Monte-Carlo methods (NeurIPS 2020 · similarity 0.89)
- ACTIVE-o3 : Empowering MLLMs with Active Perception via Pure Reinforcement Learning (ICML 2026 · similarity 0.88)
- SEEA-R1: Tree-Structured Reinforcement Fine-Tuning for Self-Evolving Embodied Agents (NeurIPS 2025 · similarity 0.86)
- DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026 · similarity 0.86)
- Act to See, See to Act: Diffusion-Driven Perception-Action Interplay for Adaptive Policies (NeurIPS 2025 · similarity 0.85)
- RLZero: Direct Policy Inference from Language Without In-Domain Supervision (NeurIPS 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.
