Risk-Sensitivity in Multi-Armed Bandits

Course Project: Multi-Armed Bandits (Monisha Jegadeesan, Shreyas Chaudhari)

The following risk-measures and algorithms are implemented for the multi-armed bandit setting:

Mean (Risk-Neutral)

• Upper Confidence Bound (UCB)

Mean-Variance (mean_variance.ipynb)

• Mean Variance - Lower Confidence Bound (MV-LCB)
  
• Exploration-Exploitation (ExpExp)
  
• Eliminative Mean Variance - Upper Confidence Bound (Eliminative MV-UCB)

Variance (var.ipynb)

• Variance - Upper Confidence Bound (VaR - UCB)
  
• Variance - Explore-then-Commit (VaR - ETC)

cVAR (cvar.ipynb)

• Multi-Armed Risk-Aware Bandit (MARAB)
  
• Multi-Armed Risk-Aware Bandit OUThandled (MARABOUT)
  
• cVAR-ETC (Conditional Variance - Explore-then-Commit)

A detailed report on the analysis of these measures and algorithms can be found here.