Speaker:

Martin Takac, Postdoctoral Fellow, Department of Computer Science

Title:

Construction of Meaning

Location:

Archway 2 - 1:00 pm, Friday 24 April

Abstract:

How do we make meaning of our experience in the world? How could artificial agents make meaning of their "experience"?

A constructivist or developmental approach to building intelligent systems emphasizes the role of incremental and continuous learning. Instead of an inbuilt (pre-programmed) knowledge/ontology, the systems are endowed with learning capabilities that allow them to acquire/construct representation of knowledge relevant to their goals during their lifetime (run-time). This approach is especially suitable for tasks in open and dynamic environments (virtual or real) with characteristics that cannot be exhaustively captured at design-time.

In my talk, I will present a semantic framework based on so-called distinguishing criteria that represent meanings of different types - objects, properties, relations, changes, situations and events, along with several computational models of their autonomous acquisition:

1. sensorimotor exploration, whereby categories are defined by common motor programs (affordances),
2. social learning focused on the influence of naming on concept formation
3. learning from episodes describing both performed and observed actions by "who did what to whom and with what result" theta role frames.