Adversarial Multi Class Learning under Weak Supervision with Performance Guarantees
Authors: Alessio Mazzetto, Cyrus Cousins, Dylan Sam, Stephen H Bach, Eli Upfal
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of our approach with experiments on various image classification tasks. Our experiments show that our method compares favorably with the recently-published ALL and PGMV algorithms for (binary classification) from weak supervision sources. |
| Researcher Affiliation | Academia | 1Department of Computer Science, Brown University. |
| Pseudocode | Yes | Algorithm 1 Subgradient Algorithm |
| Open Source Code | Yes | The code for the experiments is available online.1 1https://github.com/Bats Research/ mazzetto-icml21-code |
| Open Datasets | Yes | We demonstrate the applicability and performance of our method on image multiclass classification tasks derived from the Domain Net (Peng et al., 2019) dataset. We also provide experiments on image binary classification tasks derived from the Animals with Attributes 2 (Xian et al., 2018) dataset in order to compare our methods with additional baselines. |
| Dataset Splits | No | The paper states that weak classifiers are trained using '60% of the labeled data' and that 'm L i.i.d. labeled samples' and 'm unlabeled data points' are used. However, it does not provide specific train/validation/test splits (e.g., percentages or counts) for the main models (AMCL-CC, AMCL-LR) beyond this general description of data availability. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions using a 'pretrained Res Net-18 network' but does not specify software versions for libraries, frameworks, or programming languages (e.g., PyTorch version, Python version). |
| Experiment Setup | No | The paper mentions 'fine-tuning a pretrained Res Net-18 network' and using '60% of the labeled data' for training the weak supervision sources. However, it does not provide concrete hyperparameter values (e.g., learning rate, batch size, number of epochs) or other system-level training configurations for the main AMCL-CC and AMCL-LR models. |