Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Semantic Probabilistic Layers for Neuro-Symbolic Learning
Authors: Kareem Ahmed, Stefano Teso, Kai-Wei Chang, Guy Van den Broeck, Antonio Vergari
NeurIPS 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically demonstrate that SPLs outperform these competitors in terms of accuracy on challenging SOP tasks including hierarchical multi-label classification, pathfinding and preference learning, while retaining perfect constraint satisfaction. |
| Researcher Affiliation | Academia | Kareem Ahmed CS Department UCLA EMAIL Stefano Teso CIMe C and DISI University of Trento EMAIL Kai-Wei Chang CS Department UCLA EMAIL Guy Van den Broeck CS Department UCLA EMAIL Antonio Vergari School of Informatics University of Edinburgh EMAIL |
| Pseudocode | No | No pseudocode or algorithm block was found. |
| Open Source Code | Yes | Our code is made publicly available on Github at github.com/Kareem Yousrii/SPL. |
| Open Datasets | Yes | We use preference ranking data over 10 types of sushi for 5, 000 individuals, taken from [49], and a 60/20/20 split. |
| Dataset Splits | Yes | We use preference ranking data over 10 types of sushi for 5, 000 individuals, taken from [49], and a 60/20/20 split. |
| Hardware Specification | No | No specific hardware details (e.g., exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) were provided for running its experiments. |
| Software Dependencies | No | No specific software dependencies with version numbers were listed. The paper only mentions 'Py Torch [54]' without a version. |
| Experiment Setup | Yes | We used the validation splits to determine the number of layers in the gating function as well as the overparameterization, keeping all other hyperparameters fixed. The final models were obtained by training using a batch size of 128 and early stopping on the validation set. |