The Elusive Ground Truth: An Ongoing Journey in Data Labeling - Session by Jovan Cicvaric
Автор: MLCon | Community & Conferences
Загружено: 2025-12-16
Просмотров: 2
Описание: Data labeling is one of the most underappreciated, yet critical, stages of the machine learning lifecycle. This talk recounts our dynamic journey building a data labeling process from the ground up, a path that took us from me as the sole annotator, to outsourcing, and finally to a hybrid model. Each iteration revealed new complexities, exposing how inherent subjectivity and hidden biases challenge label consistency, even among experts. This talk moves beyond the technical to tell a story about chasing a ground truth that may not exist. We will share key learnings on managing evolving label schemas and confronting the human subjectivity that defines the process, showing how acknowledging these challenges leads to more robust machine learning.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке: