I am new to contextual MAB problems so I have a few doubts. I am trying to implement the EXP4 algorithm, however in EXP4 we only choose one arm at a time based on the advice of the Experts. We then update the weights using exponential update rule based on the loss incurred. What happens to the update rule if we have multiple arms instead of a single arm? Also, does the update rule change if I have reward instead of loss.