Abstract: Tackling unfairness is a challenging task with extensive difficulties in the context of graph
learning models. One of the major issues is posed by the absence of node attributes, due to
missing data or privacy concerns. A recent work by Guo et al. (2023) titled "Fair attribute
completion on a graph with missing attributes", tackles this problem by introducing FairAC.
The framework’s main components adopt state-of-the-art approaches, including a sensitive
discriminator and an attention mechanism to provide a solution to both the unfairness and
attribute completion problem. Supported by an experimental analysis, FairAC claims to
exhibit superior fairness performance while achieving similar node classification performance
compared to other baseline methods. In our work, we try to reproduce the results provided
by the authors along with validating their main claims. On top of that, this analysis
highlights FairAC’s ability to handle graphs with varying sparsity and fill missing attributes,
even in cases of limited neighbouring data.
Submission Length: Regular submission (no more than 12 pages of main content)
Previous TMLR Submission Url: https://openreview.net/forum?id=cRb6Zierij
Changes Since Last Submission: There were some minor discrepancies in the format of our submission e.g. header missing, "anonymous authors" and "paper under double-blinded review" lines missing. These were added and the new submission adheres to the provided TMLR latex format.
Assigned Action Editor: ~Matthew_J._Holland1
Submission Number: 2230
Loading