Research map / Graph learning / Knowledge and temporal graphs
SalKG: Learning From Knowledge Graph Explanations for Commonsense Reasoning
Aaron Chan, Jiashu Xu, Boyuan Long, Soumya Sanyal, Tanishq Gupta, Xiang Ren
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 |
|---|---|---|---|
| use ensemble-based coarse explanations instead of Grad (ablation) | SalKG-Coarse · CSQA dev, BERT+MHGRN | accuracy (%): 56.84 -> 59.49 | Improved |
| use ensemble-based coarse explanations instead of Occl (ablation) | SalKG-Coarse · CSQA dev, BERT+MHGRN | accuracy (%): 57.60 -> 59.49 | Improved |
| soft attention pruning instead of random hard pruning (ablation) | SalKG-Fine · CSQA dev, BERT+MHGRN | accuracy (%): 50.61 -> 57.28 | Improved |
| soft attention pruning instead of heuristic hard pruning (ablation) | SalKG-Fine · CSQA dev, BERT+MHGRN | accuracy (%): 50.72 -> 57.28 | Improved |
| train saliency with KL divergence instead of BCE loss (ablation) | SalKG-Fine · CSQA dev, BERT+MHGRN | accuracy (%): 50.83 -> 57.28 | Improved |
| use Occl instead of Grad for fine explanations (ablation) | SalKG-Fine · CSQA dev, BERT+MHGRN | accuracy (%): 56.05 -> 57.28 | Improved |
| use Occl instead of Grad for fine explanations (ablation) | SalKG-Hybrid · CSQA dev, BERT+MHGRN | accuracy (%): 60.17 -> 59.92 | Hurt |
| add SalKG-Coarse to KG model | KG model · CSQA test, BERT+RN | accuracy (%): 56.46 -> 57.50 | Improved |
| add SalKG-Hybrid to KG model | KG model · CSQA test, RoBERTa+RN | accuracy (%): 71.37 -> 74.13 | Improved |
| add SalKG-Coarse to KG model | KG model · OBQA test, RoBERTa+RN | accuracy (%): 66.80 -> 71.29 | Improved |
| add SalKG-Fine to KG model | KG model · OBQA test, RoBERTa+PathGen | accuracy (%): 68.40 -> 51.51 | Hurt |
| add SalKG-Hybrid to KG model | KG model · CODAH test, RoBERTa+MHGRN | accuracy (%): 84.02 -> 85.17 | Improved |
Closest papers by meaning
- (Comet-) Atomic 2020: On Symbolic and Neural Commonsense Knowledge Graphs (AAAI 2021 · similarity 0.89)
- Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward (CVPR 2026 · similarity 0.86)
- Amplifying Diversity and Quality in Commonsense Knowledge Graph Completion (Student Abstract) (AAAI 2024 · similarity 0.86)
- GraphVis: Boosting LLMs with Visual Knowledge Graph Integration (NeurIPS 2024 · similarity 0.86)
- Explainable Saliency: Articulating Reasoning with Contextual Prioritization (CVPR 2025 · similarity 0.86)
- MDK12-Bench: A Multi-Discipline Benchmark for Evaluating Reasoning in Multimodal Large Language Models (AAAI 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.
