Efficient Active Learning of Halfspaces via Query Synthesis
Authors: Ibrahim Alabdulmohsin, Xin Gao, Xiangliang Zhang
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimentally, it exhibits a significant improvement over traditional approaches such as uncertainty sampling and representative sampling. |
| Researcher Affiliation | Academia | King Abdullah University of Science & Technology (KAUST) Thuwal, Saudi Arabia 23955 |
| Pseudocode | Yes | Algorithm 1: Query synthesis algorithm for halfspaces. Data: Observations {(xi, yi)}i=1,2,..., t Result: k synthetic queries {xt+1, xt+2, . . . , xt+k} Begin: 1. Solve the optimization problem in (5) or (6). Let µ and Σ be the optimal solutions. 2. Compute N, which is the orthonormal basis to the orthogonal complement of µ (the null-space of µ T ). 3. Compute α1, α2, . . . , αk, which are the top k eigenvectors of the matrix N T Σ N. 4. Return xt+1 = Nα1, . . . , xt+k = Nαk. End |
| Open Source Code | No | MATLAB implementation codes will be made available at http://mine.kaust.edu.sa/Pages/Software.aspx |
| Open Datasets | No | The experiments involved generating data: 'For a fixed choice of k and d, we began with a random choice of a unit-norm w Rd, a single positive example, and a single negative example.' No concrete access information for a publicly available or open dataset was provided. |
| Dataset Splits | No | The paper describes an iterative query synthesis process and sample complexity, but does not specify a traditional training/validation/test dataset split or cross-validation setup. |
| Hardware Specification | No | No specific hardware details (like GPU/CPU models, memory, or cloud instances) were mentioned for the experiments. |
| Software Dependencies | Yes | The optimization problem was solved using CVX (CVX Research 2012). CVX: Matlab software for disciplined convex programming, version 2.0. Additionally, linear SVM was implemented using the LIBLINEAR package (Fan et al. 2008). |
| Experiment Setup | Yes | For a fixed choice of k and d, we began with a random choice of a unit-norm w Rd, a single positive example, and a single negative example. After that, we ran the different query synthesis algorithms in parallel up to a total of 1, 000 queries. In the batch setting, we used k = 5. Also, all experiments were repeated for d {25, 50, 75}. |