Abstract | ||
---|---|---|
For online multitask learning (oMTL), when a chunk of tasks consisting of multiple related instances is received in one batch, the learner normally has the chance to actively order these tasks to improve the learning efficiency. This paper proposes a quadratic ordering method for active oMTL, where instance ordering is integrated into task ordering by taking each instance in one task. The proposed task and instance quadratic ordering is able to facilitate oMTL better than single task ordering. The orderings derived in this paper can be incorporated into any individual oMTL algorithms for active oMTL. The performance evaluations on four real-word datasets demonstrate the benefits of the proposed algorithms. |
Year | Venue | Field |
---|---|---|
2018 | ICONIP | Multi-task learning,Computer science,Quadratic equation,Artificial intelligence,Machine learning |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
6 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Jing Zhao | 1 | 107 | 59.16 |
Shaoning Pang | 2 | 711 | 52.69 |
Iman Tabatabaei Ardekani | 3 | 14 | 7.61 |
Yuji Sekiya | 4 | 25 | 9.50 |
Daisuke Miyamoto | 5 | 32 | 4.20 |