Research map / Label-efficient and robust learning / Recognition and long-tail
A Semantic Loss Function for Deep Learning with Symbolic Knowledge
Jingyi Xu, Zilu Zhang, Tal Friedman, Yitao Liang, Guy Broeck
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 exactly-one semantic loss to an MLP | MNIST semi-supervised classification with 100 labels · MNIST, 100 labels | accuracy (%): 78.46 ($\pm$1.94) -> 98.38 ($\pm$0.51) | Improved |
| Add exactly-one semantic loss to an MLP | MNIST semi-supervised classification with 1000 labels · MNIST, 1000 labels | accuracy (%): 94.26 ($\pm$0.31) -> 98.78 ($\pm$0.17) | Improved |
| Add exactly-one semantic loss to an MLP | MNIST fully supervised classification · MNIST, all labels | accuracy (%): 99.34 ($\pm 0.08$) -> 99.36 ($\pm$0.02) | No clear effect |
| Add exactly-one semantic loss to an MLP | FASHION semi-supervised classification · FASHION, 100 labels | accuracy (%): 69.45 ($\pm 2.03$) -> 86.74 ($\pm 0.71) | Improved |
| Add exactly-one semantic loss to an MLP | FASHION semi-supervised classification · FASHION, 500 labels | accuracy (%): 78.12 ($\pm 1.41) -> 89.49 ($\pm 0.24) | Improved |
| Add exactly-one semantic loss to an MLP | FASHION fully supervised classification · FASHION, all labels | accuracy (%): 89.87 -> 89.81 | No clear effect |
| Add exactly-one semantic loss to a CNN | CIFAR-10 semi-supervised classification · CIFAR-10, 4000 labels | accuracy (%): 77.13 -> 81.79 | Improved |
| Add exactly-one semantic loss to a CNN | CIFAR-10 fully supervised classification · CIFAR-10, all labels | accuracy (%): 90.96 -> 90.92 | No clear effect |
| Add semantic loss for valid simple paths | 5-layer MLP on grid shortest paths · Grid shortest path | coherent accuracy (%): 5.62 -> 28.51 | Improved |
| Add semantic loss for valid simple paths | 5-layer MLP on grid shortest paths · Grid shortest path | incoherent accuracy (%): 85.91 -> 83.14 | Hurt |
| Add semantic loss for valid simple paths | 5-layer MLP on grid shortest paths · Grid shortest path | constraint accuracy (%): 6.99 -> 69.89 | Improved |
| Add semantic loss for valid total orderings | 3-layer MLP on sushi preference prediction · Sushi preference prediction | coherent accuracy (%): 1.01 -> 13.59 | Improved |
Closest papers by meaning
- Embedding Symbolic Knowledge into Deep Networks (NeurIPS 2019 · similarity 0.87)
- Injecting Logical Constraints into Neural Networks via Straight-Through Estimators (ICML 2022 · similarity 0.87)
- Neuro-symbolic Learning Yielding Logical Constraints (NeurIPS 2023 · similarity 0.86)
- A method to automatically discover symbolic local learning rules (NeurIPS 2026 · similarity 0.86)
- Deciphering Raw Data in Neuro-Symbolic Learning with Provable Guarantees (AAAI 2024 · similarity 0.85)
- Verification Learning: Make Unsupervised Neuro-Symbolic System Feasible (ICML 2025 · 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.
