Learning from eXtreme Bandit Feedback
Authors: Romain Lopez, Inderjit S. Dhillon, Michael I. Jordan8732-8740
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our approach on real-world datasets with a supervised learning to bandit feedback conversion (Dudik, Langford, and Li 2011; Gentile and Orabona 2014). We report results on three datasets from the Extreme Classification Repository (Bhatia et al. 2016), with L ranging from several thousand to half a million (Table 1). Table 2: Performance comparisons of POXM and other competing methods over the three medium-scale datasets. All experiments are conducted with bandit feedback. |
| Researcher Affiliation | Collaboration | Romain Lopez1, Inderjit S. Dhillon2, 3, Michael I. Jordan1, 2 1 Department of Electrical Engineering and Computer Sciences, University of California, Berkeley 2 Amazon.com 3 Department of Computer Science, The University of Texas at Austin |
| Pseudocode | No | The paper describes algorithmic procedures but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions that its implementation relies on PyTorch and refers to an arXiv link for supplementary information ('Please visit https://arxiv.org/abs/2009.12947 for supplementary information.'), but does not state that its own source code is released or provide a repository link for its methodology. |
| Open Datasets | Yes | We report results on three datasets from the Extreme Classification Repository (Bhatia et al. 2016)... URL http://manikvarma.org/downloads/XC/XMLRepository.html. |
| Dataset Splits | No | Table 1 provides 'Ntrain: #training instances, Ntest: #test instances' and states 'The partition of training and test is from the data source.', giving explicit counts for training and test sets. However, there is no explicit mention or details about a separate validation dataset split. |
| Hardware Specification | Yes | In this study, we used a machine with 8 GPUs Tesla K80 to run our experiments. |
| Software Dependencies | No | The paper mentions 'Py Torch' as the implementation framework but does not provide specific version numbers for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | Furthermore, hyperparameters that are specific to Attention XML are fixed across all experiments (Table 2 of You et al. (2019b))... Consequently, we lowered the learning rate for these algorithms from 1e 4 to 5e 5... we focus on the grid {0.7,0.8,0.9,1.0}. |