Algorithms for Optimizing the Ratio of Submodular Functions

Authors: Wenruo Bai, Rishabh Iyer, Kai Wei, Jeff Bilmes

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we empirically demonstrate the performance and good scalability properties of our algorithms. ... 6. Experiments: We empirically evaluate our proposed algorithmic frameworks for RS minimization, including in particular MMIN, GREEDRATIO, and ELLIPSOIDAPPROX, on a synthetic data set.
Researcher Affiliation Academia Wenruo Bai WRBAI@UW.EDU Rishabh Iyer RKIYER@U.WASHINGTON.EDU Kai Wei WEIKAI.HUST@GMAIL.COM Jeff Bilmes BILMES@UW.EDU University of Washington, Seattle, WA 98195, USA
Pseudocode Yes Algorithm 1 A (1 + ϵ)-approximation algorithm for RS minimization using an exact algorithm for DS minimization. ... Algorithm 2 Approx. algorithm for RS minimization using an approximation algorithm for SCSC. ... Algorithm 3 GREEDRATIO for RS minimization. ... Algorithm 4 ELLIPSOIDAPPROX. ... Algorithm 5 MMIN for RS minimization.
Open Source Code No The paper does not provide any statement or link indicating that open-source code for the described methodology is available.
Open Datasets No We empirically evaluate our proposed algorithmic frameworks for RS minimization, including in particular MMIN, GREEDRATIO, and ELLIPSOIDAPPROX, on a synthetic data set. In particular, we evaluate on a generalized form of the F-measure function: Fλ(X) = |Γ(X) T| λ|T| + (1 λ)|Γ(X)|, (32) where 0 λ 1 is a parameter that determines a tradeoff weight between precision and recall. Note Fλ=0.5 is the same as the F-measure function defined in Eqn. 3. We instantiate the F-measure function on a randomly generated bipartite graph G(U, W, E). The bipartite graph is defined with |U| = 100 and |W| = 100. We define an edge between u U and w W independently with probability p = 0.05. The set of targets T W is also randomly chosen with a fixed size 20, i.e., |T| = 20. We run the experiments on 10 instances of the randomly and independently generated data, and we report the averaged results.
Dataset Splits No The paper describes the generation of synthetic data for evaluation but does not specify distinct training, validation, or test splits for a model or algorithm, as the algorithms themselves are being evaluated rather than a trained model.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments.
Software Dependencies No The paper does not provide any specific software dependencies with version numbers.
Experiment Setup Yes We instantiate the F-measure function on a randomly generated bipartite graph G(U, W, E). The bipartite graph is defined with |U| = 100 and |W| = 100. We define an edge between u U and w W independently with probability p = 0.05. The set of targets T W is also randomly chosen with a fixed size 20, i.e., |T| = 20. We run the experiments on 10 instances of the randomly and independently generated data, and we report the averaged results. ... As a baseline, we also implement a random sampling method, where we randomly choose 100 subsets X U with size |X| = 50 and report their averaged function valuation in terms of Fλ and their standard deviation. In Figure 1, we compare the performance of all methods on with the varying λ.