Research map / Large language models / Language, safety and interpretability
CLEVRER-Humans: Describing Physical and Causal Events the Human Way
Jiayuan Mao, Xuelin Yang, Xikun Zhang, Noah Goodman, Jiajun Wu
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 rule-based pairwise event detector before event generation (ablation) | pair description generator · stage III human labels, 100 videos | pairwise description accuracy: 80.6% to 87.2% | Improved |
| pretrain on CLEVRER then test on CLEVRER-Humans | ALOE · CLEVRER-Humans test | per-option accuracy: 98.5 -> 54.0 | Hurt |
| finetune pretrained model on CLEVRER-Humans (ablation) | ALOE · CLEVRER-Humans test | per-option accuracy: 54.0 -> 52.7 | No clear effect |
| train from scratch vs pretrain+finetune (ablation) | ALOE · CLEVRER-Humans test | per-question accuracy: 31.7 -> 32.1 | No clear effect |
| replace LSTM with pretrained BERT | CNN+LSTM · CLEVRER-Humans test | per-option accuracy: 50.3 -> 52.9 | No clear effect |
| train CNN+LSTM from scratch on CLEVRER-Humans (ablation) | CNN+LSTM · CLEVRER-Humans test | per-question accuracy: 30.0 -> 34.2 | Mixed |
| use CLEVRER heuristic instead of counterfactual labels to approximate humans | causal labels · CLEVRER sample | agreement with human | Improved |
| train longer on small training set (ablation) | ALOE · CLEVRER-Humans test | per-option accuracy: 70% train; 53.7% test | No clear effect |
Closest papers by meaning
- COLD: Causal reasOning in cLosed Daily activities (NeurIPS 2024 · similarity 0.86)
- ReXTime: A Benchmark Suite for Reasoning-Across-Time in Videos (NeurIPS 2024 · similarity 0.85)
- VidEvent: A Large Dataset for Understanding Dynamic Evolution of Events in Videos (AAAI 2025 · similarity 0.84)
- Two Causally Related Needles in a Video Haystack (NeurIPS 2025 · similarity 0.84)
- MECD: Unlocking Multi-Event Causal Discovery in Video Reasoning (NeurIPS 2024 · similarity 0.84)
- EventLens: Event-Structure Reinforcement Learning for Video Understanding (NeurIPS 2026 · 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.
