HyperLogic: Enhancing Diversity and Accuracy in Rule Learning with HyperNets
Authors: Yang Yang, Wendi Ren, Shuang Li
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on real data demonstrate that our method can learn more diverse, accurate, and concise rules. Our code is publicly available at https://github.com/Yang Yang-624/Hyper Logic. |
| Researcher Affiliation | Academia | Yang Yang1, Wendi Ren1, Shuang Li1 1School of Data Science, The Chinese University of Hong Kong (Shenzhen) {yangyang8, wendiren}@link.cuhk.edu.cn, lishuang@cuhk.edu.cn |
| Pseudocode | No | The paper describes its methods in prose and mathematical equations but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is publicly available at https://github.com/Yang Yang-624/Hyper Logic. |
| Open Datasets | Yes | We selected four publicly available binary classification datasets: MAGIC gamma telescope (magic), adult census (adult), FICO HELOC (heloc), and home price prediction (house). In all datasets, preprocessing was performed to encode categorical and numerical attributes as binary variables, which can be found in [17]. |
| Dataset Splits | Yes | Table 1: Test accuracy based on a nested 5-fold cross-validation (%, mean standard error). ... We then select the rule set with the highest training accuracy as the optimal set |
| Hardware Specification | Yes | For our method, all experiments were conducted on a Linux server with an Intel(R) Xeon(R) Gold 6248R CPU @ 3.00GHz and 30Gi of memory, running Ubuntu 20.04.5 LTS, using one of the NVIDIA Ge Force RTX 3090 GPUs available on the server. |
| Software Dependencies | No | The paper mentions "Ubuntu 20.04.5 LTS" as the operating system and "Adam as the optimizer" but does not specify version numbers for other key software libraries or frameworks (e.g., Python, PyTorch, TensorFlow). |
| Experiment Setup | Yes | We use Adam as the optimizer, and the learning rate is 1e-4, with weight decay is 1e-4. The number of training epochs is 10000. In the experiments for selecting the optimal rules for comparison, we set hyperparameter M1 = 5, M2 = 5000, λ1 = 0.01, and λ2 = 0.1 (λ1 and λ2 are related to the hypernetwork loss or regularization loss, as defined in Eq. (6)). |