Guarantees for Greedy Maximization of Non-submodular Functions with Applications
Authors: Andrew An Bian, Joachim M. Buhmann, Andreas Krause, Sebastian Tschiatschek
ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We experimentally validate our theoretical findings for both synthetic and real-world applications. |
| Researcher Affiliation | Academia | 1Department of Computer Science, ETH Zurich, Zurich, Switzerland. |
| Pseudocode | Yes | Algorithm 1: The GREEDY Algorithm |
| Open Source Code | Yes | Source code is available at https://github.com/bianan/non-submodular-max. |
| Open Datasets | Yes | We used the Boston Housing Data. The dataset6 has 14 features (e.g., crime rate, property tax rates, etc.) and 516 samples. (Footnote 6: https://archive.ics.uci.edu/ml/datasets/ Housing) ... For real-world data, we considered an active set selection task on the CIFAR-107 dataset. (Footnote 7: https://www.cs.toronto.edu/ kriz/cifar. html) |
| Dataset Splits | No | The paper mentions using datasets like Boston Housing and CIFAR-10 but does not provide specific details on how these datasets are split into training, validation, and test sets, or if standard splits were used without explicit reference. |
| Hardware Specification | No | The paper states that experiments were implemented using Matlab and an SDP solver, but it does not provide specific details about the hardware (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments. |
| Software Dependencies | Yes | All experiments were implemented using Matlab. We used the SDP solver provided by CVX (Version 2.1). |
| Experiment Setup | Yes | In all experiments, we normalized the data points to have unit 2-norm. ... For synthetic data, we generated random covariance matrices 2 Rn n with uniformly distributed eigenvalues in [0, 1]. We set n = 10, σ = 2. ... For real-world data, we considered an active set selection task on the CIFAR-10 dataset. ... squared exponential kernel (k(xi, xj) = exp( kxi xjk2/h2), h was set to be 1). |