Task Detectors for Progressive Systems

Maxwell Joseph Jacobson, Purdue University

Abstract

While methods like learning-without-forgetting [1] and elastic weight consolidation [2] accomplish high-quality transfer learning while mitigating catastrophic forgetting, progressive techniques such as Deepmind’s progressive neural network accomplish this while completely nullifying forgetting. However, progressive systems like this strictly require task labels during test time. In this paper, I introduce a novel task recognizer built from anomaly detection autoencoders that is capable of detecting the nature of the required task from input data. Alongside a progressive neural network or other progressive learning system, this task-aware network is capable of operating without task labels during run time while maintaining any catastrophic forgetting reduction measures implemented by the task model.

Degree

M.Sc.

Advisors

Rodriguez-Rivera, Purdue University.

Subject Area

Artificial intelligence

COinS