Limits on Testing Structural Changes in Ising Models
Authors: Aditya Gangrade, Bobak Nazer, Venkatesh Saligrama
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Fig. 3 illustrates Thm. 7 via a simulation for testing deletions in a binary tree (for p = 127, α = 0.1), showing excellent agreement. In particular, observe the sharp drop in samples needed at s = 21 2 p versus at s < p 11. We note that SL-based testing fails for all s 60 for this setting even with 1500 samples (Fig. 4 in C.3), which is far beyond the scale of Fig. 3. See C.3 for details. |
| Researcher Affiliation | Academia | Aditya Gangrade Boston University gangrade@bu.edu Bobak Nazer Boston University bobak@bu.edu Venkatesh Saligrama Boston University srv@bu.edu |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or link to open-source code for the methodology described in the paper. |
| Open Datasets | No | The paper uses samples generated from a p = 127 node binary tree with α = 0.1 for its simulations, which is synthetic data generated by the authors, not a publicly available dataset. |
| Dataset Splits | No | The paper describes simulations but does not specify train/validation/test dataset splits. |
| Hardware Specification | No | The paper states it used Python and NumPy for Monte-Carlo simulations but does not provide specific details about the hardware used (e.g., CPU, GPU models, memory). |
| Software Dependencies | Yes | Code for verifying bounds for forest-structured Ising models: We used Python 3.7.3 (conda 4.8.3) and NumPy 1.17.4 to run Monte-Carlo simulations. |
| Experiment Setup | Yes | For the results in Figure 3, we used samples generated from a p = 127 node binary tree with α = 0.1. The simulation was run for n = 200 samples and repeated for 5000 independent trials. Each trial estimates the risk over 100 independent pairs of (P, Q) chosen from the null and alternate hypothesis spaces, and the risk for each (P, Q) pair is computed by comparing the statistic T with a threshold chosen by cross-validation over 10 independent pairs of (P, Q). |