Credal Self-Supervised Learning
Authors: Julian Lienen, Eyke Hüllermeier
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In an exhaustive empirical evaluation, we compare our methodology to state-of-the-art self-supervision approaches, showing competitive to superior performance especially in low-label scenarios incorporating a high degree of uncertainty. |
| Researcher Affiliation | Academia | Julian Lienen Department of Computer Science Paderborn University Paderborn 33098, Germany julian.lienen@upb.de Eyke Hüllermeier Institute of Informatics University of Munich (LMU) Munich 80538, Germany eyke@ifi.lmu.de |
| Pseudocode | Yes | The pseudo-code of the complete algorithm can be found in the appendix. |
| Open Source Code | No | The paper does not include an explicit statement about releasing source code or provide a link to a code repository. |
| Open Datasets | Yes | More precisely, we follow the semi-supervised learning evaluation setup as described in [41] and perform experiments on CIFAR-10/-100 [24], SVHN [34], and STL-10 [8] with varying fractions of labeled instances sampled from the original data sets, also considering label-scarce settings with only a few labels per class. |
| Dataset Splits | No | The paper states it performs experiments on standard datasets like CIFAR-10 and SVHN using 'varying fractions of labeled instances' and refers to an external paper for the 'semi-supervised learning evaluation setup', but it does not explicitly provide the specific percentages or sample counts for training, validation, or test splits within the text provided. |
| Hardware Specification | No | The paper mentions 'computing time provided by the Paderborn Center for Parallel Computing (PC2)' but does not specify any particular hardware details such as GPU models, CPU types, or memory used for the experiments. |
| Software Dependencies | No | The paper mentions software components like CTAugment, SGD with Nesterov momentum, and cosine annealing, but does not provide specific version numbers for any of these or other software dependencies. |
| Experiment Setup | Yes | To guarantee a fair comparison to existing methods related to Fix Match, we keep the hyperparameters the same as in the original experiments. ... we set the learning rate to η cos 7πk / 16K , where η is the initial learning rate, k the current training step and K the total number of steps (2^20 by default). |