Ambiguity-Aware Abductive Learning
Authors: Hao-Yuan He, Hui Sun, Zheng Xie, Ming Li
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we conduct experiments to verify our claims and validate the superior performance of A3BL. Specifically, the experiments focus on two challenging tasks in the neurosymbolic field: Digit Addition and Handwritten Formula Recognition. To ensure reproducibility, all experiments are repeated five times, each with a different random seed. |
| Researcher Affiliation | Academia | 1National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China 2School of Artificial Intelligence, Nanjing University, Nanjing, China. Correspondence to: Ming Li <lim@lamda.nju.edu.cn>. |
| Pseudocode | Yes | The pseudo-code of A3BL can be referred in Algorithm 1. |
| Open Source Code | No | The paper states that implementations of compared methods are based on their official packages and provides links to those, but it does not explicitly state that the source code for A3BL itself is open-source or provided: 'The implementation of ABL-hamming and ABL-conf is based on their official package. The implementation of NGS is based on their official package. The implementation of Deep Stoch Log is based on their official package. The implementation of Deep Prob Log is based on their official package. The implementation of Neur ASP is based on their official package.' |
| Open Datasets | Yes | Building upon this concept, we have expanded the task to incorporate four distinct datasets: MNIST (Deng, 2012), KMNIST (Clanuwat et al., 2018), CIFAR10 (Krizhevsky, 2009), and SVHN (Netzer et al., 2011). |
| Dataset Splits | No | The paper explicitly states the size of the training sets for different tasks, for instance, 'Digit Addition, n=1 2 30000 10000' where 30000 is the number of sequences in the training set and 10000 in the test set. However, it does not specify explicit validation splits (e.g., percentages, counts, or dedicated validation sets). |
| Hardware Specification | Yes | The results were obtained using a computer setup consisting of an Intel Xeon Platinum 8538 CPU and an NVIDIA A100PCIE-40GB GPU on an Ubuntu 20.04 Focal platform. All experiments were conducted on a system equipped with an NVIDIA Ge Force RTX 3090 GPU, Intel Xeon Silver 4210 CPU, 64GB of RAM, and Ubuntu 20.04 Focal. |
| Software Dependencies | No | The paper mentions using 'Adam (Kingma & Ba, 2015) as the optimizer' but does not specify version numbers for programming languages, libraries, or frameworks used for implementation. |
| Experiment Setup | Yes | Table 3 provides detailed experimental setup parameters: 'Method Optimizer Learning rate Batch size Epoch ... A3BL Adam 0.001 256 50 Adam 0.001 1024 50'. |