Graph-Aided Online Multi-Kernel Learning
Authors: Pouya M. Ghari, Yanning Shen
JMLR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on a number of real datasets showcase the advantages of our novel graph-aided algorithms compared to state-of-the-art alternatives. ... This section presents experimental results over real datasets downloaded from UCI Machine Learning Repository (Dua and Graff, 2017). The accuracy of different approaches are evaluated using mean square error (MSE). |
| Researcher Affiliation | Academia | Pouya M. Ghari EMAIL Department of Electrical Engineering and Computer Science University of California Irvine, CA 92697, USA; Yanning Shen EMAIL Department of Electrical Engineering and Computer Science University of California Irvine, CA 92697, USA |
| Pseudocode | Yes | Algorithm 1 Data Driven (Bipartite) Graph-based Kernel Selection ... Algorithm 2 Generating Bipartite Feedback Graph ... Algorithm 3 OMKL with (Bipartite) Graph Feedback (OMKL-GF) ... Algorithm 4 Generating Similarity based Feedback Graph ... Algorithm 5 OMKL with Similarity-based Feedback Graph (OMKL-SFG) ... Algorithm 6 OMKL with Similarity Feedback Graph Refinement (OMKL-SFG-R) |
| Open Source Code | Yes | Codes are available at https://github.com/pouyamghari/Graph-Aided-Online-Multi-Kernel-Learning. |
| Open Datasets | Yes | This section presents experimental results over real datasets downloaded from UCI Machine Learning Repository (Dua and Graff, 2017). The performance of kernel learning algorithms is evaluated through several real datasets: Airfoil... Bias... Concrete... Naval... |
| Dataset Splits | No | The paper describes an online learning setting where data samples {(xt, yt)}T t=1 are collected in a sequential fashion. It evaluates performance over time on real datasets but does not specify traditional training/test/validation splits with percentages or absolute counts for reproducibility in a batch learning sense. The evaluation metrics like MSE are calculated cumulatively over time (e.g., sum from tau=1 to t of (f(x_tau)-y_tau)^2). |
| Hardware Specification | Yes | All experiments were carried out using Intel(R) Core(TM) i7-10510U CPU @ 1.80 GHz 2.30 GHz processor with a 64-bit Windows operating system. |
| Software Dependencies | No | The paper mentions a general '64-bit Windows operating system' but does not provide specific version numbers for programming languages, libraries, or other software dependencies used in the experiments. |
| Experiment Setup | Yes | The number of random features is D = 50. The kernel dictionary contains 76 kernels including 51 RBF kernels and 25 Laplacian kernels. The bandwidth of the i-th (1 i 51) RBF kernel is 10σi with σi = 2i 52 / 25. And the value of the i-th (1 i 25) Laplacian kernel s parameter is 10λi where λi = i 13 / 6. For fairness of evaluation, parameters ξ, η and ηe are set to 0.1 / t for all algorithms at time step t. Parameter λ is set to 10^-3 for all proposed algorithms OMKL-GF, OMKL-SFG and OMKL-SFG-R. The maximum number of kernels chosen by OMKL-GF at each time instant is 10. In addition, for OMKL-SFG, to determine the value of γi for each vertex vi V, the number of out-neighbors for each node is set to be 10. For OMKL-SFG-R, at each time, β is set to β = (1 ξ) u[10,t] + ξ / N where u[10,t] denote the tenth greatest value in the sequence {ui,t / Ut }N i=1. |