Minimax regret with unbounded weights

HIGHLIGHTS

• What: In contrast the authors show in this paper that for unbounded norm with R ∞ the minimax regret asymptotically (d - 1) log(T /d) √for logistic-like expert class which the authors also extend to R Ω( T ). The authors focus on precise asymptotics of regrets rT* and r̄T* for large d, T, and unbounded weights w. In Theorem 4 the authors demonstrate that for logistic function, an (d - 3) log(T /d) lower bound is achievable even with √ R=Ω( T ).
• Who: Michael Drmota and colleagues from the Vienna, Austria Purdue University have published the . . .
  

   

  

  If you want to have access to all the content you need to log in!

  Thanks :)

  Username or E-mail:

  If you don't have an account, you can create one here.