Lifelong Machine Learning Test
Lianghao Li, Qiang Yang · 2015
In this paper, we propose to measure the intelligence of an agent by measuring how fast its knowledge lev-el increases after learning related tasks. In this paper, we propose a new Lifelong Machine Learning Test. An agent can pass the test if it can learn unrestricted num-ber of tasks over time and its knowledge level can in-crease when new tasks are learned. In the proposed test, both an agent’s current performance and its per-formance growth rate are taking into account. We also give a new theoretical requirement and a novel empiri-cal evaluation metric for the proposed lifelong machine learning test.