Showing 1–2 of 2 results for author: Holstege, F
-
Removing Spurious Concepts from Neural Network Representations via Joint Subspace Estimation
Authors:
Floris Holstege,
Bram Wouters,
Noud van Giersbergen,
Cees Diks
Abstract:
Out-of-distribution generalization in neural networks is often hampered by spurious correlations. A common strategy is to mitigate this by removing spurious concepts from the neural network representation of the data. Existing concept-removal methods tend to be overzealous by inadvertently eliminating features associated with the main task of the model, thereby harming model performance. We propos…
▽ More
Out-of-distribution generalization in neural networks is often hampered by spurious correlations. A common strategy is to mitigate this by removing spurious concepts from the neural network representation of the data. Existing concept-removal methods tend to be overzealous by inadvertently eliminating features associated with the main task of the model, thereby harming model performance. We propose an iterative algorithm that separates spurious from main-task concepts by jointly identifying two low-dimensional orthogonal subspaces in the neural network representation. We evaluate the algorithm on benchmark datasets for computer vision (Waterbirds, CelebA) and natural language processing (MultiNLI), and show that it outperforms existing concept removal methods
△ Less
Submitted 18 October, 2023;
originally announced October 2023.
-
Female scholars need to achieve more for equal public recognition
Authors:
Menno H. Schellekens,
Floris Holstege,
Taha Yasseri
Abstract:
Different kinds of "gender gap" have been reported in different walks of the scientific life, almost always favouring male scientists over females. In this work, for the first time, we present a large-scale empirical analysis to ask whether female scientists with the same level of scientific accomplishment are as likely as males to be recognised. We particularly focus on Wikipedia, the open online…
▽ More
Different kinds of "gender gap" have been reported in different walks of the scientific life, almost always favouring male scientists over females. In this work, for the first time, we present a large-scale empirical analysis to ask whether female scientists with the same level of scientific accomplishment are as likely as males to be recognised. We particularly focus on Wikipedia, the open online encyclopedia that its open nature allows us to have a proxy of community recognition. We calculate the probability of appearing on Wikipedia as a scientist for both male and female scholars in three different fields. We find that women in Physics, Economics and Philosophy are considerable less likely than men to be recognised on Wikipedia across all levels of achievement.
△ Less
Submitted 16 April, 2019; v1 submitted 12 April, 2019;
originally announced April 2019.
