Research map / Probabilistic and scientific modeling / Dynamical systems
A Compositional Object-Based Approach to Learning Physical Dynamics
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 |
|---|---|---|---|
| use pairwise factorization and compose pair interactions | NPE · Four-ball prediction task | velocity prediction | Helped |
| add a neighborhood mask to select context objects (ablation) | NPE · Four-ball prediction task | velocity prediction loss: about an order of magnitude improvement | Helped |
| add a neighborhood mask to select context objects (ablation) | NPE · Generalization to variable object count | velocity prediction loss: about an order of magnitude improvement | Helped |
| remove the neighborhood mask (ablation) | NPE · Mass inference | accuracy | Hurt |
| set the neighborhood threshold within the range 3 to 5 ball radii (ablation) | NPE · Constant-mass prediction task | velocity prediction: N(o_f) \in [3, 5] | No clear effect |
| use a neighborhood threshold smaller or larger than the robust range (ablation) | NPE · Constant-mass prediction task | velocity prediction | Hurt |
| use longer input durations than two timesteps (ablation) | NPE · Physics prediction | prediction performance | No clear effect |
| generalize with the compositional NPE to unseen object counts | NPE · Generalization to 6, 7, and 8 balls | velocity prediction: 0.5 to 1 order of magnitude | Helped |
| use the NPE rather than the NP or LSTM | NPE · Different scene configurations | cosine similarity of the velocity angle | Helped |
| use the NPE rather than the NP or LSTM | NPE · Different scene configurations | relative error in magnitude | Helped |
Closest papers by meaning
- ComPhy: Compositional Physical Reasoning of Objects and Events from Videos (ICLR 2022 · similarity 0.89)
- Interaction Networks for Learning about Objects, Relations and Physics (NeurIPS 2016 · similarity 0.88)
- Neural Force Field: Few-shot Learning of Generalized Physical Reasoning (ICLR 2026 · similarity 0.88)
- Reasoning About Physical Interactions with Object-Oriented Prediction and Planning (ICLR 2019 · similarity 0.87)
- NeuMA: Neural Material Adaptor for Visual Grounding of Intrinsic Dynamics (NeurIPS 2024 · similarity 0.86)
- Universal Physics Transformers: A Framework For Efficiently Scaling Neural Operators (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.