Horospherical Decision Boundaries for Large Margin Classification in Hyperbolic Space
Authors: Xiran Fan, Chun-Hao Yang, Baba Vemuri
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present several experiments depicting the competitive performance of our classifier in comparison to SOTA. In this section, we present several experimental results obtained from an application of our Horo SVM to synthetic data as well as real data sets used in published literature. |
| Researcher Affiliation | Academia | Xiran Fan Department of Statistics University of Florida fanxiran@ufl.edu Chun-Hao Yang Institute of Statistics and Data Science National Taiwan University chunhaoy@ntu.edu.tw Baba C. Vemuri Department of CISE University of Florida vemuri@ufl.edu |
| Pseudocode | No | The paper describes its methods mathematically and textually but does not include any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions that its implementation is based on Pymanopt [19] but does not provide a link or explicit statement about releasing its own source code for the described methodology. |
| Open Datasets | Yes | Here, we follow the experimental setup in [12], and evaluate our Horo SVM over four real-world network data sets used by [8]: karate [34] (2 classes, 34 nodes ), polblogs [2] (2 classes, 1224 nodes ), polbooks 3 (3 classes, 105 nodes ), and football [17] (12 classes, 115 nodes ). We obtained hyperbolic embeddings in various dimensions using the approach in [15] for Word Net 4 noun hierarchy (82,115 nodes). |
| Dataset Splits | Yes | We conducted a five-fold cross-validation on each data set, where we chose the hyperparameter C from {1, 5, 10} during the cross-validation procedure. We split all nodes in a subtree into positive training (80%) and test (20%) nodes and applied the same process to the remaining Word Net nodes to create negative training and test sets. We split the dataset into training and test sets with 100 positive/negative samples in the training set and 100 positive/negative samples in the test set. |
| Hardware Specification | Yes | Our implementation is based on Pymanopt [19] using the Riemannian conjugate gradient method [26] on Intel(R) Xeon(R) CPU E5-2683 v3 @ 2.00GHz. |
| Software Dependencies | No | The paper mentions 'Pymanopt [19]' but does not provide specific version numbers for Pymanopt or any other software dependencies. |
| Experiment Setup | Yes | We conducted a five-fold cross-validation on each data set, where we chose the hyperparameter C from {1, 5, 10} during the cross-validation procedure. We chose the hyperparameter C from {1, 5, 10} during the cross-validation procedure. The average training times for each method on one dataset (200 samples) as follows: 6.57 seconds (Hyperboloid SVM), 3.98 seconds (Hyperbolic LR), 9.06 seconds (HNN), and 3.73 seconds (Horo SVM). |