Dear Students,

Today, we had our last formal lecture of the topics to be covered of our
textbook.  And thus, this is also the last installment of thoughts for this
semester.  Here are some thoughts:

1.  Learning and Communication are symbiotic, especially in a MAS.  Agents can
     learn to improve their communication; and agents communicate to learn.

2.  Agents can improve their communication in four aspects.  What are these?
     Why is it important to improve communication in each of these aspects?  If
     somebody gives you a domain problem to solve, will you be able to decide
     which aspects will be improved based on multiagent learning, and which
     aspects will be addressed through "pre-fabrication"? That is, which
     aspects are suitable for just being part of the original agent design to
     address; which aspects are suitable for the agent to learn to address?

3.  Agents need to communicate to learn from or about each other.  There are
     two general levels: low and high.  Low-level communication is
     data/information sharing.  High-level communication is
     understanding/reasoning sharing.  In the MAS research community, data
     sharing is not always considered learning.  Why?

     Further, low-level communication is easier to design and implement
     compared to high-level communication.  Thus, when one designs a MAS that
     relies on communication to learn, one should also consider the cost of
     learning.  High-level communication could usually lead to more insightful
     "learning", but at a higher cost.

We have thus concluded our formal lectures.  We will have presentations on
advanced topics next week.

Thank you for your attention,

LeenKiat