Online Learning of Multiple Tasks and Their Relationships : Testing on Spam Email Data and EEG Signals Recorded in Construction Fields
Yixin Jin, Wenjing Zhou, Meiqi Wang, Meng Li, Xintao Li, Tianyu Hu
TL;DR
The paper addresses the limitation of assuming fixed task relatedness in online multitask learning by introducing a dynamic framework that jointly adapts task weights $w$ and the interaction matrix $A$ using online updates. It presents three update rules, $OMTLCOV$, $OMTLLOG$, and $OMTLVON$, and contrasts them with a fixed $A$ method (CMTL). Through spam and construction workers’ EEG datasets, the approach shows that adaptive relatedness can yield improvements in EEG classification accuracy by about $1$–$3\%$ and maintains competitive spam error rates around $12\%$, with EEG tasks benefiting most from the dynamic coupling. The work demonstrates practical potential for real-time, cross-task learning in data-rich, time-sequential settings, while highlighting the importance of epoch scheduling, learning rate, and feature quality for robust performance.
Abstract
This paper examines an online multi-task learning (OMTL) method, which processes data sequentially to predict labels across related tasks. The framework learns task weights and their relatedness concurrently. Unlike previous models that assumed static task relatedness, our approach treats tasks as initially independent, updating their relatedness iteratively using newly calculated weight vectors. We introduced three rules to update the task relatedness matrix: OMTLCOV, OMTLLOG, and OMTLVON, and compared them against a conventional method (CMTL) that uses a fixed relatedness value. Performance evaluations on three datasets a spam dataset and two EEG datasets from construction workers under varying conditions demonstrated that our OMTL methods outperform CMTL, improving accuracy by 1% to 3% on EEG data, and maintaining low error rates around 12% on the spam dataset.