Tilburg center for Cognition and Communication (TiCC)

We study how people communicate with each other and how computer systems can be taught to communicate with us.

vignetweb

TiCC Colloquium: George Kachergis

What: Learning More with Less Input: Lessons from Children and Education for A.I. ‒ and vice-versa!
Where: TZ 3
When: Wednesday, 8 November 2017, 12:45 - 13:45 hours


Abstract

Learning in animals is a slow process, in which the order of experiences often has a large effect on what is learned. This order-dependent learning is generally seen as undesirable in a machine learning context, where many Bayesian (i.e., 'rational') models are viewed normatively ‒ casting animal (and human) learning as irrational; suboptimal. In this talk, I will discuss several major differences between the contexts of animal and machine learning (ML), and argue that our learning system is adapted quite well ‒ and far better than ML, at the moment ‒ for the problems we face. In particular, I will examine the problem of language acquisition, considering the amount of input humans need (hint: less than ML), as well as the order and origin of that data, as it is actively selected both by learners and teachers. I suggest that in order to create artificial intelligence that understands language in a similar way to us, we need to investigate and understand human learning behavior, and see why it is adaptive in its context. Throughout, I will mention research at the intersection of human and machine learning that aims to give us this understanding, and how even now we can create AI-driven applications based on what we know to improve (and study) education.

About George Kachergis

Website

When: 08 November 2017 12:45

End date: 08 November 2017 13:45