Bipartite Ranking From Multiple Labels: On Loss Versus Label Aggregation

Authors: Michal Lukasik, Lin Chen, Harikrishna Narasimhan, Aditya Krishna Menon, Wittawat Jitkrittum, Felix X. Yu, Sashank J. Reddi, Gang Fu, Mohammadhossein Bateni, Sanjiv Kumar

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present a suite of synthetic and real-world experiments to empirically validate our theoretical findings.
Researcher Affiliation Industry 1Google Research. Correspondence to: Michal Lukasik <EMAIL>, Lin Chen <EMAIL>.
Pseudocode No The paper describes methods and analyses them, but does not provide any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statement about providing source code or a link to a code repository.
Open Datasets Yes We consider the UCI Banking dataset composed of information about Bank customers, advertising campaign details, and the success thereof (Moro et al., 2014). Help Steer (Wang et al., 2023b) consists of evaluations of LLM responses across 5 categories: MSLR. We next consider MSLR Web30k, a dataset of users query-document interactions (Qin & Liu, 2013).
Dataset Splits Yes We train a 3-layer MLP model with hidden dimension 256 and Re LU activation over 8K examples for 50 epochs. We evaluate on held out 2K examples.
Hardware Specification No The paper does not provide specific hardware details used for running its experiments.
Software Dependencies No The paper mentions using the 'TensorFlow ranking library' but does not specify any software versions for libraries or programming languages.
Experiment Setup Yes We train a 3-layer MLP model with hidden dimension 256 and Re LU activation over 8K examples for 50 epochs. We train a linear model on numerical features using Adam for 100 epochs.