Assisted Labeling Visualizer (ALVI): A Semi-Automatic Labeling System For Time-Series Data

Abstract

Machine learning applications can significantly benefit from large amounts of labeled data, although the task of labeling data is notoriously challenging and time-consuming. This is particularly evident in domains involving human subjects, where labeling time-series signals often necessitates trained professionals. In this work, we introduce the Assisted Labeling Visualizer (ALVI), a system that simplifies the process of labeling data by offering an interactive user interface that visualizes synchronized video, feature-map representations, and raw time-series signals. ALVI also leverages deep learning and self-supervised learning techniques to facilitate the semi-automatic labeling of large amounts of unlabeled data. We demonstrate the capabilities of ALVI on a human activity recognition dataset to showcase its potential for enhancing the labeling process of time-series sensor data.