Adapting Kernel Representations Online Using Submodular Maximization
Authors: Matthew Schlegel, Yangchen Pan, Jiecao Chen, Martha White
ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically demonstrate the effectiveness of this approximation, in terms of approximation quality, significant runtime improvements, and effective prediction performance. ... We empirically illustrate the accuracy and efficiency of our proposed method as compared to Online Greedy with no approximation to the submodular function (which we call Full Greedy), Sieve Streaming, and various naive versions of our algorithm. We also show this method can achieve reasonable regression accuracy as compared with KRLS (Engel et al., 2004). For these experiments we use four well known datasets: Boston Housing (Lichman, 2015), Parkinson s Telemonitoring (Tsanas et al., 2010), Sante Fe A (Weigend, 1994) and Census 1990 (Lichman, 2015). |
| Researcher Affiliation | Academia | 1Department of Computer Science, Indiana University, Bloomington. Correspondence to: Martha White <martha@indiana.edu>. |
| Pseudocode | Yes | Algorithm 1 Online Greedy Algorithm 2 Block Greedy: Online Greedy for Prototype Selection using a Block-Diagonal Approximation Algorithm 3 Block Greedy-Swap(x) |
| Open Source Code | No | The paper does not contain an explicit statement about the release of source code or a link to a code repository. |
| Open Datasets | Yes | For these experiments we use four well known datasets: Boston Housing (Lichman, 2015), Parkinson s Telemonitoring (Tsanas et al., 2010), Sante Fe A (Weigend, 1994) and Census 1990 (Lichman, 2015). |
| Dataset Splits | No | The paper mentions training on a portion of the data (e.g., 'train on the first 1000 time steps') and evaluating with cross-validation ('averaged over 50 runs') but does not specify explicit train/validation/test splits by percentage or count for all datasets. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for the experiments (e.g., GPU/CPU models, memory specifications). |
| Software Dependencies | No | The paper does not provide specific version numbers for software dependencies or libraries used in the experiments. |
| Experiment Setup | Yes | Input: threshold parameter t, where a prototype is only added if there is sufficient improvement ... For all experiments on maximization quality, we use percentage gain with a threshold of t = 0.001. ... We set the budget size to b = 80, which is smaller than what KRLS used, and chose a block size of 4. ... We used a budget of b = 500, and a block size of r = 25... For all regression experiments, we use a threshold of t = 0.01 unless otherwise specified. ... The regularizer λ = 0.001, and the utility threshold is t = 0.0001. |