Feature Density
Feature density, the concentration of data points in a feature space, is a burgeoning research area impacting various machine learning tasks. Current research focuses on leveraging feature density for improved uncertainty estimation, particularly in out-of-distribution detection, and for enhancing model performance in applications like sign language translation and semantic segmentation, often employing techniques like normalizing flows and contrastive learning. Understanding and manipulating feature density offers potential for more efficient and robust machine learning models, reducing computational costs and improving accuracy across diverse domains.
Papers
Exploring the Potential of Feature Density in Estimating Machine Learning Classifier Performance with Application to Cyberbullying Detection
Juuso Eronen, Michal Ptaszynski, Fumito Masui, Gniewosz Leliwa, Michal Wroczynski
Initial Study into Application of Feature Density and Linguistically-backed Embedding to Improve Machine Learning-based Cyberbullying Detection
Juuso Eronen, Michal Ptaszynski, Fumito Masui, Gniewosz Leliwa, Michal Wroczynski, Mateusz Piech, Aleksander Smywinski-Pohl