Active learning and graphical models

Andrea Bertozzi
University of California, Los Angeles (UCLA)

One of the challenges for machine learning methods is the lack of training
data. Active learning refers to methods in which a "human in the loop"
is part of the algorithm. Graphical models are a useful mathematical bridge between the theory of autonomous agent-based systems in control theory and machine learning methods. Graph structure is important for understanding the dynamics of consensus models and it is also central to developing accurate methods for learning algorithms using similarity graphs.
Embedding a human in such methods is challenging, especially in terms of designing a method to efficiently use automation to incorporate the resources of the human.
I will review some background on similarity graphs, graph Laplacians and
how to use their structure for semi-supervised learning. I will discuss
current work on active learning using this approach.

Back to Workshop IV: Social Dynamics beyond Vehicle Autonomy