Dear all,
you are invited to the thesis defense of Jan Martin on the 21st of December, 14:00.
The topic will be "Applying Structural Analogy to Solve Abstract Reasoning Problems in a More Human-like Way" with the below given abstract.
We will use the link for our virtual colloquium which you can find in this VC course.
Best regards
Johannes
Abstract:
Creating actual Intelligence, an entity capable of creativity and reason, is an implicit goal of Artificial Intelligence by its very name. The ability to learn, abstracting knowledge from one domain and applying it to a different domain, is a core fundamental of human reasoning that allows learning and understanding a new concept with just a few examples, by virtue of having already learned from examples of other, similar domains. It stands to reason that teaching a machine those abilities will play a pivotal role in creating Artificial General Intelligence, and the measurement of those abilities will be a requirement to evaluating this.
This Thesis aims to investigate the viability of applying the Copycat architecture, built to solve 1-dimensional analogies, to 2-dimensional tasks taken from a subset of the Abstract Reasoning Corpus, as used in the related Abstract Reasoning Challenge. Examples will be picked based on complexity, as well as manually chosen, fed into Metacat, a more advanced development of Copycat, and evaluated in regards to both their current output as well as their theoretical potential.