Toward Controlling Discrimination in Online Ad Auctions

Authors: Elisa Celis, Anay Mehrotra, Nisheeth Vishnoi

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical results on the A1 Yahoo! dataset demonstrate that our algorithm can obtain uniform coverage across different user types for each advertiser at a minor loss to the revenue of the platform, and a small change to the size of the audience each advertiser reaches.
Researcher Affiliation Academia L. Elisa Celis 1 Yale University, USA Anay Mehrotra 2 Indian Institute of Technology Kanpur, India Nisheeth K. Vishnoi 3 Yale University, USA.
Pseudocode Yes Algorithm 1 Algorithm1(Q, G, L, η, µmax, µmin, ε)
Open Source Code No The paper does not provide an explicit statement or link for the release of its source code.
Open Datasets Yes We use the Yahoo! A1 dataset (Yahoo), which contains bids placed by advertisers on the top 1000 keywords on Yahoo! Online Auctions between June 15, 2002 and June 14, 2003.
Dataset Splits No The paper uses the Yahoo! A1 dataset and mentions running experiments over iterations but does not specify train, validation, or test dataset splits.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments.
Software Dependencies No The paper mentions general software like 'Scikit-learn' and 'SciPy' in its references, but does not specify versions for these or other software dependencies used in their experiments.
Experiment Setup Yes We set ℓi1 = ℓi2 = ℓ i [2], and vary ℓuniformly from 0 to 0.5 , i.e., from the completely unconstrained case (which is equivalent to Myerson’s action) to completely constrained case (which requires each advertiser to win each keywords in the pair with exactly the same probability). We report κN,M, d T V (N, M), and slift(F) averaged over all auctions after 10^4 iterations in Figure 1;