Large-scale Multi-label Learning with Missing Labels

Authors: Hsiang-Fu Yu, Prateek Jain, Purushottam Kar, Inderjit Dhillon

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we present extensive empirical results on a variety of benchmark datasets and show that our methods perform significantly better than existing label compression based methods and can scale up to very large datasets such as a Wikipedia dataset that has more than 200,000 labels.
Researcher Affiliation Collaboration Hsiang-Fu Yu ROFUYU@CS.UTEXAS.EDU Department of Computer Science, University of Texas at Austin Prateek Jain PRAJAIN@MICROSOFT.COM Purushottam Kar T-PURKAR@MICROSOFT.COM Microsoft Research India, Bangalore Inderjit S. Dhillon INDERJIT@CS.UTEXAS.EDU Department of Computer Science, University of Texas at Austin
Pseudocode Yes Algorithm 1 General Loss with Missing Labels... Algorithm 2 Squared Loss with Full Labels
Open Source Code No The paper does not include an unambiguous statement or a direct link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We considered a variety of benchmark datasets including four standard datasets (bibtex, delicious, eurlex, and nus-wide), two datasets with d L (autofood and compphys), and a very large scale Wikipedia based dataset, which contains about 1M wikipages and 200K labels. See Table 1 for more information about the datasets. [...] Finally, we show the scalability of our method by applying it to a recently curated Wikipedia dataset (Agrawal et al., 2013), that has 881,805 training samples and 213,707 labels.
Dataset Splits No The paper mentions 'Training set' and 'Test set' in Table 1 but does not explicitly specify the methodology or percentages for splitting data into training, validation, and test sets to ensure reproducibility beyond general dataset names.
Hardware Specification Yes We conducted all experiments on an Intel machine with 32 cores.
Software Dependencies No The paper mentions software tools like LIBLINEAR and TRON, but it does not provide specific version numbers for these or any other ancillary software components required for replication.
Experiment Setup No While the paper describes algorithmic choices and reports the number of alternating iterations (five), it does not explicitly provide specific hyperparameter values (e.g., learning rates, batch sizes, regularization strengths) or detailed training configurations for its own methods to enable reproduction.