Research map / Multi-agent systems and games / Human-AI and goal-conditioned
CompILE: Compositional Imitation Learning and Execution
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 unsupervised CompILE segmentation instead of LSTM surprisal | Reacher boundary prediction · Reacher, 3 tasks | accuracy: 24.8 \pm 0.6 -> 62.0 \pm 4.5 | Helped |
| use unsupervised CompILE segmentation instead of LSTM surprisal | Reacher boundary prediction · Reacher, 5 tasks | accuracy: 21.6 \pm 0.5 -> 41.7 \pm 8.0 | Helped |
| add supervision on segment encodings z | CompILE · Reacher, 3 tasks | accuracy: 62.0 \pm 4.5 -> 99.5 \pm 0.2 | Helped |
| add supervision on boundary positions | CompILE · Reacher, 3 tasks | accuracy: 62.0 \pm 4.5 -> 99.8 \pm 0.1 | Helped |
| add supervision on segment encodings z | CompILE · Reacher, 5 tasks | accuracy: 41.7 \pm 8.0 -> 98.4 \pm 0.5 | Helped |
| add supervision on boundary positions | CompILE · Reacher, 5 tasks | accuracy: 41.7 \pm 8.0 -> 98.8 \pm 0.3 | Helped |
| replace discrete codes with 32-dimensional Gaussian latent variables (ablation) | CompILE · Reacher, 3 tasks | accuracy: 62.0 \pm 4.5 -> 45.2 \pm 13.8 | Hurt |
| replace discrete codes with 32-dimensional Gaussian latent variables (ablation) | CompILE · Reacher, 5 tasks | accuracy: 41.7 \pm 8.0 -> 28.7 \pm 7.0 | Hurt |
| increase the allowed number of segments from 3 to 4 (ablation) | CompILE · Reacher, 3 tasks | accuracy: 62.0 \pm 4.5 -> 53.4 \pm 6.3 | Hurt |
| increase the allowed number of segments from 3 to 5 (ablation) | CompILE · Reacher, 3 tasks | accuracy: 62.0 \pm 4.5 -> 29.8 \pm 6.1 | Hurt |
| lower the fixed Gumbel softmax temperature to 0.1 (ablation) | CompILE · Reacher, 3 tasks | accuracy: 62.0 \pm 4.5 -> 40.9 \pm 5.3 | Hurt |
| lower the fixed Gumbel softmax temperature to 0.01 (ablation) | CompILE · Reacher, 3 tasks | accuracy: 62.0 \pm 4.5 -> 36.3 \pm 6.0 | Hurt |
Closest papers by meaning
- Interactive and Hybrid Imitation Learning: Provably Beating Behavior Cloning (NeurIPS 2025 · similarity 0.88)
- CraftFactory: A Conditioned Control Policy Benchmark for Compositional Generalization (AAAI 2025 · similarity 0.87)
- Quantifying Generalisation in Imitation Learning (NeurIPS 2025 · similarity 0.87)
- Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations (NeurIPS 2025 · similarity 0.87)
- Data Augmentation for Instruction Following Policies via Trajectory Segmentation (AAAI 2025 · similarity 0.87)
- Temporal Representation Alignment: Successor Features Enable Emergent Compositionality in Robot Instruction Following (NeurIPS 2025 · similarity 0.87)
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.