Files in this item



application/pdfECE499-Sp2018-lin.pdf (837kB)Restricted to U of Illinois
(no description provided)PDF


Title:Language acquisition and object recognition with Bert
Author(s):Lin, Yuguang
Contributor(s):Levinson, Stephen
Subject(s):Artificial Intelligence
Machine Learning
Gaussian Mixture Model
Hidden Markov Model
Abstract:Recent advances in the broad field of artificial intelligence (AI) has brought much excitement and many expectations. However, there is a strong need to understand intelligence, and through understanding it we can help achieve true machine intelligence, one that is not only able to complete certain difficult tasks but also reason about the world. To study intelligence, we look at ourselves and especially at infants. At a very young age, we can consciously and easily perform tasks that involve understanding of both languages and vision, two of the channels through which we acquire most of our information from the external world. How do we do so? How do children learn their first language? In our lab, we believe intelligence and learning should be interactive. We learn from interaction with the real world through our five senses. We also believe a massive number of well-defined labels does not exist for children. In order to study this idea of unlabeled data and learning through interaction, for this thesis we implemented a system that enabled a human robot to associate visual information with speech information and to learn to describe a new object with vocabularies acquired during training. Several machine learning models were implemented on the iCub humanoid platform. Specifically, Gaussian Mixture Models and K-means were implemented for the vision part of the experiment, and Hidden Markov Models were used for the speech.
Issue Date:2018-05
Date Available in IDEALS:2018-05-24

This item appears in the following Collection(s)

Item Statistics