Sum of Ranked Range Loss for Supervised Learning
Authors: Shu Hu, Yiming Ying, Xin Wang, Siwei Lyu
JMLR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our empirical results highlight the effectiveness of the proposed optimization frameworks and demonstrate the applicability of proposed losses using synthetic and real data sets. We empirically demonstrate the robustness and effectiveness of the proposed Ao RR, TKML, TKML-Ao RR, and their optimization frameworks on both synthetic and real data sets. |
| Researcher Affiliation | Academia | Shu Hu EMAIL Department of Computer Science and Engineering University at Buffalo, State University of New York Buffalo, NY 14260-2500, USA Yiming Ying EMAIL Department of Mathematics and Statistics University at Albany, State University of New York Albany, NY 12222, USA Xin Wang EMAIL Department of Computer Science and Engineering University at Buffalo, State University of New York Buffalo, NY 14260-2500, USA Siwei Lyu EMAIL Department of Computer Science and Engineering University at Buffalo, State University of New York Buffalo, NY 14260-2500, USA |
| Pseudocode | Yes | Algorithm 1: DCA for Minimizing So RR Algorithm 2: DCA for Minimizing Ao RR without Setting k and m Algorithm 3: Combination of Ao RR and TKML |
| Open Source Code | Yes | Code available at https://github.com/discovershu/So RR. |
| Open Datasets | Yes | Our empirical results highlight the effectiveness of the proposed optimization frameworks and demonstrate the applicability of proposed losses using synthetic and real data sets. We use the MNIST data set (Le Cun et al., 1998) We use five benchmark data sets from the UCI (Dua and Graff, 2017) and the KEEL (Alcalá-Fdez et al., 2011) data repositories |
| Dataset Splits | Yes | For each data set, we first randomly select 50% samples for training, and the remaining 50% samples are randomly split for validation and testing (each contains 25% samples). To create a validation set, We randomly extract 10, 000 samples from training samples. Therefore, the remaining training data size is 50, 000. For each data set, we randomly partition it to 50%/25%/25% samples for training/validation/testing, respectively. We randomly split Yeast data into two parts, which are 80% samples for training and 20% samples for testing. |
| Hardware Specification | Yes | All algorithms are implemented in Python 3.6 and trained and tested on an Intel(R) Xeon(R) CPU W5590 @3.33GHz with 48GB of RAM. |
| Software Dependencies | Yes | All algorithms are implemented in Python 3.6 |
| Experiment Setup | Yes | Hyper-parameters C, k, and m are selected based on the validation set. Specifically, parameter C is chosen from {100, 101, 102, 103, 104, 105}, parameter k ∈ {1} ∪ [0.1 : 0.1 : 1]n, where n is the number of training samples, and parameter m is selected in the range of [1, k). Algorithm 1: DCA for Minimizing So RR Algorithm 2: DCA for Minimizing Ao RR without Setting k and m Algorithm 3: Combination of Ao RR and TKML |