Towards Unbiased Exploration in Partial Label Learning
Authors: Zsolt Zombori, Agapi Rissaki, Kristóf Szabó, Wolfgang Gatterbauer, Michael Benedikt
JMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We give a theoretical justification for our loss function, and provide an extensive evaluation of its impact on synthetic data, on standard partially labelled benchmarks and on a contributed novel benchmark related to an existing rule learning challenge. Keywords: partial label learning, disjunctive supervision, rule learning |
| Researcher Affiliation | Academia | Zsolt Zombori EMAIL Alfred R enyi Institute of Mathematics E otv os Lor and University Budapest, Hungary Agapi Rissaki EMAIL Khoury College of Computer Sciences Northeastern University Boston, USA Krist of Szab o EMAIL Alfr ed R enyi Institute of Mathematics Budapest, Hungary Wolfgang Gatterbauer EMAIL Khoury College of Computer Sciences Northeastern University Boston, USA Michael Benedikt EMAIL Department of Computer Science University of Oxford Oxford, UK |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. It primarily uses mathematical formulations and descriptive text. |
| Open Source Code | Yes | The entire codebase is available from the project webpage (BESS project 23). |
| Open Datasets | Yes | We provide novel PLL data sets appropriate for rule learning in a supervised context. The entire codebase is available from the project webpage (BESS project 23). ... We use the setup from Wen et al. (2021), starting from the CIFAR10 (Krizhevsky and Hinton, 2009) image classification benchmark ... In the following we experiment with the partially labelled rule learning data sets, introduced in Section 5. ... We experiment with the CMT challenges, described earlier in Section 5. ... The RODI data set was introduced in Pinkel et al. (2015, 2018)... The NPD challenge (Skjæveland et al., 2013). ... we adapt five real-world PLL data sets, each targeting a different task: Lost (Cour et al., 2011), Soccer Player (Zeng et al., 2013), and Yahoo!News (Guillaumin et al., 2010) for automatic face naming from video frames or images, MSRCv2 (Liu and Dietterich, 2012) for object classification and Bird Song (Briggs et al., 2012) for bird song classification. |
| Dataset Splits | Yes | All experiments employ a (70%, 15%, 15%) train-validation-test split. ... For this experiment we apply 10-fold cross validation to evaluate all losses |
| Hardware Specification | Yes | A single experiment lasts for around 7hours on a single Nvidia A100 GPU. ... A single experiment lasts for around 23 hours on a single Nvidia A100 GPU. |
| Software Dependencies | No | All experiments are performed using Pytorch. The paper mentions PyTorch but does not provide a specific version number, which is required for reproducibility. |
| Experiment Setup | Yes | We use learning rate 0.1 and weight decay with parameter 10^-3. We train for 300 epochs using Stochastic Gradient Descent with batches of size 256. All experiments are performed using Pytorch. |