"Dependence and Model Selection in LLP: The Problem of Variants" was accepted at KDD 2023.

Is there a single “best” algorithm for Learning from Label Proportions? We argue no: the dependence structure between bags, items, and labels defines distinct LLP variants, and accounting for it leads to better model selection across a wide range of datasets and algorithms.

• Paper
• Code

For a short explanation, here’s the thread:

Is there a (single) “best” algorithm for Learning from Label Proportions (LLP)? In our KDD 23 @kdd_news paper, we argue that the answer is NO. There are many variants of LLP, which have an important effect on the choice of the solution method https://t.co/Sw7LuszpkL [1/12]

— Gabriel Franco (@gvsfranco) August 5, 2023