Research map / Probabilistic and scientific modeling / Dynamical systems
Interaction Networks for Learning about Objects, Relations and Physics
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 relation-centric function phi_R (full IN vs dynamics-only IN) (ablation) | Interaction network · n-body prediction | MSE: 76 -> 0.25 | Helped |
| add phi_R (full IN vs dynamics-only IN) (ablation) | Interaction network · Balls prediction | MSE: 0.074 -> 0.0020 | Helped |
| add phi_R (full IN vs dynamics-only IN) (ablation) | Interaction network · String prediction | MSE: 0.017 -> 0.0011 | Helped |
| use IN instead of flattened-input MLP baseline | next-step velocity prediction · n-body prediction | MSE: 79 -> 0.25 | Helped |
| use IN instead of flattened-input MLP baseline | next-step velocity prediction · String prediction | MSE: 0.016 -> 0.0011 | Helped |
| use IN instead of constant velocity baseline | next-step velocity prediction · Balls prediction | MSE: 0.074 -> 0.0020 | Helped |
| use IN instead of MLP for potential energy estimation | abstract estimation · String potential energy | MSE: 425 -> 1.1 | Helped |
| use IN instead of MLP for potential energy estimation | abstract estimation · n-body potential energy | MSE: 19 -> 1.4 | Helped |
| test IN trained on 6 bodies on 3 bodies, versus training on the 3-body system (ablation) | Interaction network · n-body 3 generalization | MSE: 0.21 -> 0.052 | Helped |
| test IN trained on 6 bodies on 12 bodies, versus training on 12 bodies (ablation) | Interaction network · n-body 12 generalization | MSE: 5.2 -> 7.8 | Hurt |
| transfer to 5-mass string versus training on it (ablation) | Interaction network · String 5, 1 generalization | MSE: 0.0030 -> 0.00052 | Helped |
| transfer to other ball counts versus training on them (ablation) | Interaction network · Balls 9 generalization | MSE: 0.0019 -> 0.004 | Hurt |
Closest papers by meaning
- Graph Networks as Learnable Physics Engines for Inference and Control (ICML 2018 · similarity 0.90)
- Visual Interaction Networks: Learning a Physics Simulator from Video (NeurIPS 2017 · similarity 0.89)
- A Compositional Object-Based Approach to Learning Physical Dynamics (ICLR 2017 · similarity 0.88)
- Neural Relational Inference for Interacting Systems (ICML 2018 · similarity 0.88)
- UGPhysics: A Comprehensive Benchmark for Undergraduate Physics Reasoning with Large Language Models (ICML 2025 · similarity 0.85)
- Scaling Physical Reasoning with the PHYSICS Dataset (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.