This  talk  has  two  parts.  The  first  part  is  Bayesian  and  describes  structural  results  for  partially  observed  Markov  
decision processes in multi-agent systems when individual agents perform social learning. Two specific examples 
are considered.  The first example deals with the so called constrained optimal social learning problem where the 
onset  of  herding  is  delayed  by  agents  sharing  full  information.  The  second  example  deals  with  change  detection  
when individual agents perform social learning. 
The second part  of the talk discusses regret based stochastic approximation algorithms for  learning  correlated equilibria 
in repeated games with Markov switched parameters. The convergence analysis involves differential inclusions.