Updates are available Correction dated 2025-06-05

Article History

Received: 14 June 2022

Revised: 10 August 2023

Accepted: 17 October 2023

First Online: 27 November 2023

Change Date: 23 June 2025

Change Type: Update

Change Details: The original online version of this article was revised: The following error appeared in the paper of V. Verreet, L. De Raedt, J. Bekker, “model[1]ing PU learning using probabilistic logic programming”, which appeared in volume 113 of machine learning. Table 1 on page 14 incorrectly shows the program?:: mech ← Pos for the probabilistic gap assumption. The atom Pos should be replaced with an independent Dum.y variable Dum.that has the exact same probability as Pos, resulting in the program?:: mech ← Dum. Similarly the program for the probabilistic Bridge assumption should contain the rules?:: mech and?:: mech ← Dum. The initial programs are erroneous because the prob[1]ability for an instance with attributes X to be labeled under the probabilistic gap assumption should be P(lab | x) = M(x)C(x) = k P(pos | x) · P(pos | x) = kP(pos | x)2 In ProbLog, the first factor P(pos | x) comes from the probability that pos is true and the sec[1]ond factor from the probability that dum is true. If these probabilities are not independent, the labeling probability would incorrectly be P(lab | x)=kP(pos | x), which is the SCAR assumption with label frequency k.

Change Date: 5 June 2025

Change Type: Correction

Change Details: A Correction to this paper has been published:

Change Details: https://doi.org/10.1007/s10994-025-06790-5

Declarations

:

: There are no conflicts of interest.

: There are no ethical concerns.

: N/A.

: N/A.

Free to read: This content has been made available to all.