Nice lecture! Something to clarify, in the temperature example, the epsilon does not mean the interval has length epsilon. It means the temperature has epsilon probability landing on that interval. The actual length of the interval could be more or less than epsilon.
should equation 6) be: 2e^(-epsilon*m/2)? This is because the chance of sampling from the whole highlighted region is epsilon, so the probability of sampling from a specific region is epsilon/2? Thank you for the great lecture!
Steps are unclear. Going from 5) to 6) is doesn't seem so clear, even when 5) is correctly written. Also, not clear how can one bound with 1-epsilon the probability of x_k not being in CA. Also, going into 5), there isn't a substitution, but inequality transitivity.
Serious answer: Indeed, it assumes that the true class definition is a linear interval. Joke answer: The Coloradan doesn't cannot appreciate anything as being nice if it's flat.
Jordan you are the best lecture in teaching Machine learning theory.
Thank you! However, completely false. Rob Schapire is definitely the best teacher I've seen. Unfortunately he doesn't have many (any?) UA-cam videos.
Nice lecture! Something to clarify, in the temperature example, the epsilon does not mean the interval has length epsilon. It means the temperature has epsilon probability landing on that interval. The actual length of the interval could be more or less than epsilon.
Interesting way of presenting while teaching, never seen this method before!
should equation 6) be: 2e^(-epsilon*m/2)? This is because the chance of sampling from the whole highlighted region is epsilon, so the probability of sampling from a specific region is epsilon/2? Thank you for the great lecture!
Yes I think so too, epsilon/2 for each side
Steps are unclear. Going from 5) to 6) is doesn't seem so clear, even when 5) is correctly written. Also, not clear how can one bound with 1-epsilon the probability of x_k not being in CA. Also, going into 5), there isn't a substitution, but inequality transitivity.
equation 5 is wrong.by the way, a good lecture.
Nice lecture! But, doesn't this analysis assume that Coloradoan concept of "nice" is clustered in a linear space?
Serious answer: Indeed, it assumes that the true class definition is a linear interval.
Joke answer: The Coloradan doesn't cannot appreciate anything as being nice if it's flat.
@@JordanBoydGraber Thanks Prof!
where's the continuation of this lecture?
users.umiacs.umd.edu/~jbg/teaching/CMSC_726/
has links to all lectures
Thanks!
@@JordanBoydGraber Hi, seems like this link is no longer working. If you have an updated link it would be a huge help!
@@hamza3838 Don't know why they keep changing this!
users.umiacs.umd.edu/~jbg/teaching/CMSC_726/