Focused learning promotes continual task performance in humans

Abstract

Humans can learn to perform multiple tasks in succession over the lifespan ("continual" learning), whereas current machine learning systems fail. Here, we investigated the cognitive mechanisms that permit successful continual learning in humans. Unlike neural networks, humans that were trained on temporally autocorrelated task objectives (focussed training) learned to perform new tasks more effectively, and performed better on a later test involving randomly interleaved tasks. Analysis of error patterns suggested that focussed learning permitted the formation of factorised task representations that were protected from mutual interference. Furthermore, individuals with a strong prior tendency to represent the task space in a factorised manner enjoyed greater benefit of focussed over interleaved training. Building artificial agents that learn to factorise tasks appropriately may be a promising route to solving continual task performance in machine learning.

A neural network walks into a lab: towards using deep nets as models for human behavior

“…Along similar lines, in the realm of perceptual learning, DNNs can be tested for the same qualitative properties as observed in humans, such as increased specificity (Wenliang & Seitz, 2018). Other efforts that focused on learning dynamics have moved beyond conventional DNNs to unsupervised training (Anselmi et al, 2016;Eslami et al, 2018;Flesch, Balaguer, Dekker, Nili, & Summerfield, 2018;Lotter et al, 2017;Stoianov & Zorzi, 2012;Watanabe et al, 2018) or to architectures with strong prior knowledge (Lake et al, 2015).…”

Editorial: Perspectives for natural language processing between AI, linguistics and cognitive science

“…However, it can be argued that the move toward deep learning has the potential of bringing NLP back to its roots after all. Some recent activities and findings in this direction include: Techniques like multi-task learning have been used to integrate cognitive data as supervision in NLP tasks (Barrett et al, 2016); Pre-training/finetuning regimens are potentially interpretable in terms of cognitive mechanisms like general competencies applied to specific tasks (Flesch et al, 2018); The ability of modern models for 'few-shot' or even 'zero-shot' performance on novel tasks mirrors human performance (Srivastava et al, 2018); Evidence of unsupervised structure learning in current neural network architectures that mirrors classical linguistic structures (Hewitt and Manning, 2019;Tenney et al, 2019).…”