1You can make a ton of money in this scenario with the stock market. You try to predict if a stock will go up or down. Later, you can see if the stock went up or down.234If you run this algorithm forever, how many errors can you make?51 seems like a pretty good number to output, but C2 might be a must better classifier, so a simple majority might not be the best answer6Here, the always right expert is the neighbors dog. We start off not knowing who is always right. How can we figure out where it is?7This does not necessarily mean that we can find the always right expert in log(n) iterations8log(n) mistakes before restarting each time. How many restarts? No more than [OPT+1], where OPT is the number of mistakes the best expert makes.910How do we know halving the weight is best? Maybe some other decrease in weight is
View Full Document