Quotes from Spring 1996

Instructor quotes

• I probably won't assign many homeworks, but the ones I do assign will be long and time consuming.--1-9-96
• I don't want to embarrass you. You can raise your hand lightly, but who doesn't know what a maximum likelihood estimator is?--1-9-96
• What I am about to say is important...this is where you want to pay attention.--1-9-96
• If there was a die hard frequentest in the room, I would probably be assassinated on the spot.--1-9-96
• If you have no data, you have a problem situation.--1-9-96
• You do a little soul searching and ask yourself, "What is unsatisfying about this solution?"--1-9-96
• RV means random variable, not recreational vehicle.--1-11-96
• It's okay to use abusive notation as long as it is clear in your mind that you are being very abusive.--1-11-96
• Sometimes this point seems perfectly clear even when it isn't.--1-11-96
• I may as well apologize for being a little sloppy in what I am going to do next.--1-11-96
• Oh goodness, I'm late and I'm the first one here. [a bit of an exaggeration]--1-16-96
• I was late for class today, but I think I'm going to have to put my foot down and demand that I show up on time and demand that you do too.--1-16-96
• I'm going to write this in a non-causal way. [Writes notation and then the equation]--1-16-96
• The nice thing is that if my formula were wrong, it wouldn't really matter...I'd still get the right answer.--1-16-96
• Make this a little aside. I'm going off on a tangent.--1-16-96
• I believe that all... this is like a Steve Martin routine.--1-16-96
• I just wrote that because everybody knows that filters are written with an "h".--1-16-96
• This is a little bit of a lie, but I'm going to say it just for the sake of keeping things clear.--1-18-96
• I left off this term because it doesn't matter, but let me put it in anyway just so nobody can accuse me of cheating.--1-18-96
• That's the right answer. It looks right to me too, but it isn't the answer I wanted to get.--1-18-96
• For some of you this may be boring, but bear with me.--1-18-96
• If you go look at this program, it will make your head hurt.--1-18-96
• There are three files that participate in this program.--1-18-96
• The first announcement I want to make... what are you doing here?--1-23-96 [To one of his students who took the class a previous semester.]
• The first announcement I want to make is I'm not going to be here for awhile.--1-23-96
• There aren't enough people like us and most of them don't know what a tensor is.--1-23-96
• X, Y, Z... I'm trying to think of another letter we can use.--1-23-96
• I'm going to assign this as a problem because it's just too complicated to do in class.--1-23-96
• Oh, we are supposed to be finished now aren't we? Let me just... [We got out 8 minutes later.]--1-23-96
• This is the notation that I told you you shouldn't use, but I'm going to use it anyway.--2-6-96
• That may seem confusing, and it is, but there are historical reasons.--2-6-96
• You should have had this beaten into your head in EE600.--2-6-96
• Is that how you spell satisfy? Close enough. You can always spell it right in your notes, if you choose.--2-6-96
• This is a legacy of the bad notation we learned as undergrads.--2-6-96
• Oh come on, everyone has to vote.--2-6-96
• Of course, my vote is right since I cheated and knew the answer.--2-6-96
• There was a time when the signal processing community went along with this because it seemed fair, but it's wrong.--2-6-96
• Let me check that this is right before I start arguing that it is.--2-7-96
• All the way through you're going to see a method, a method known as ignoring the constant.--2-7-96
• There are many image restoration algorithms that are much worse than doing nothing. In fact, we'll do some in class.--2-7-96
• It's instructive to see how optimal something can be and still be bad.--2-7-96
• Notice that my notation is very abusive... but I'm not going to bother to fix it.--2-7-96
• Well let me just write down the answer from this piece of paper.--2-7-96
• ...ICM which means something that means very little to you.--2-7-96
• You have 5 knobs on your TV, usually two of which you don't know what they do.--2-7-96
• So what I want to do is... what I want to do is put a box around it because it's useful.--2-8-96
• If I start being too precise, my notation is going to get miserable.--2-8-96
• I'm really being abusive with the notation, but it's too much trouble to do it correctly.--2-8-96
• I'm sorry I spontaneously changed the definition of K.--2-13-96
• We could train little kids to listen to speech backwards and then parse it.--2-13-96
• We could run the whole world backwards.--2-13-96
• Do not read this paper. This is the most complicated paper I've ever read. If anyone can read this and explain it to me, they will get an automatic A in this class. [He then proceeded to have us use a portion of this paper on our take home midterm exam.]--2-13-96
• [In the above paper] He proves a result that is almost miraculous, but nearly useless.--2-13-96
• What's 9 times 4? [Student: 36] Thanks.--2-15-96
• What's the cost of L(1,3)?... well it's not quite clear from the problem statement I gave you.--2-15-96
• Computation is of order MN... let me erase that [erases MN]. What do you think the computation is the order of?--2-15-96
• If you have to wait until the end of time before processing data on your computer, you've got a problem.--2-15-96
• This is an interesting problem, but it's far enough away from my field that I'm not sure how interesting it is.--2-15-96
• Here you need on the order of the amount of time it takes for the sun to burn up.--2-15-96
• If this were true, life would be good. We wouldn't be teaching this class, or it would be at the undergrad level because it would actually be useful, and I would be out of a job.--2-15-96
• This is like waiting for the universe to end.--2-15-96
• This is a big improvement over that, but it's still useless.--2-15-96
• Class is actually supposed to end at quarter after. I may actually end on time. [Stated at 10:16. Class dismissed at 10:20.]--2-15-96
• You are not supposed to talk to any human or any other advanced life form about anything technical during this 24 hour period. [During our take home exam]--2-20-96
• During the take home exam you should isolate yourself from all EE, statisticians, etc...--2-20-96
• If I find out you cheated, you will be answerable to me, and I will take it personally forever.--2-20-96
• Before I go any further I have to define a bunch of junk.--2-20-96
• I'm using the word lattice because it makes everyone feel more comfortable, but it's really just a set.--2-20-96
• S can be the table leg, clock, and pull cord if you want.--2-20-96
• Historically this is the terminology that was used, so blame it on the physicists since they came up with it.--2-20-96
• It's like that saying, 'You can't always get what you want, but if you try hard enough, you can get what you need. [or something like that]' I shouldn't be quoting the Rolling Stones since they did that Microsoft commercial.--2-20-96
• That was a writo... it made no sense... I must have been dreaming about something.--2-22-96
• This reminds me of those GRE questions where they say, 'Sue is sitting next to some guy who is ten feet tall who is...' and you're supposed to find yourself. You think 'Argh, why are they asking me this?'--2-22-96
• If they were really cruel on the GRE's they'd make it so there wasn't an answer or the answer was that you had to be sitting by yourself.--2-22-96
• If there were enough letters in the alphabet, we could do this.--2-22-96
• This notation is a little perverse, but it will work for this proof.--2-22-96
• This is sort of a non-causal explanation.--2-22-96
• You can go on and lead a normal life without ever understanding this proof... but it would be nice to be one of the few, the proud who understand this proof, even though you will never need to know it.--2-22-96
• An ellipse of dots... that's an interesting concept.--2-22-96
• This proof is like taking a balloon and squeezing it. You've got to make sure it doesn't pop out somewhere else.--2-22-96
• I won't torture you any longer with this. I'll sweep the rest under the carpet and say, 'By induction...'--2-22-96
• It might be that all math is broken, but assuming that math is true, all this is true too.--2-22-96
• I'm not going to do what I said I was going to do.--2-26-96
• There's ambient energy, that's what keeps all the molecules in this room from falling to the floor.--2-26-96
• Everyone should take a statistical thermodynamics course because you can prove all of the laws of nature from scratch.--2-26-96
• Okay, there was a point I wanted to make here... amazingly enough.--2-26-96
• Let me do sort of a boring calculation just to show you a calculation with this stuff.--2-26-96
• Let me just leave it now and then you can come back and yell at me later.--2-26-96
• This is a very hard problem. There is an analytic solution to this... it's in that paper I told you not to read.--2-26-96
• I'm going to have to try to wing this. Don't write this down yet.--2-26-96
• It's probably a little contrived since I found four iterations that I liked.--2-27-96
• I'll put an exclamation point there because it's supposed to be surprising.--2-27-96
• I'll copy a section of the book since it would be somewhat inappropriate to hand out photocopies of the whole book.--2-27-96
• I could go into a long spiel on why this is called ICM, but let me just say it is...--2-27-96
• The thing about this is that most of the Markov Random Fields that people use are not proper, but they ignore that minor detail.--2-27-96
• You keep doing this until your tired and then...--3-12-96
• This is like, to flagrantly use a physical analogy, is like force.--3-19-96
• This is like the peer pressure function. It tells how much the pixel is influenced by its neighbors.--3-19-96
• You can go through the proof and all that, but if it weren't true, there would be something very wrong.--3-19-96
• You run this until you get tired of waiting for the computer, and then you'll be pretty confident you have the right answer.--3-19-96
• On our last project we did image restoration. You probably noticed that the restored images didn't look any better.--3-21-96
• I didn't want to give the impression that this stuff actually works.--3-21-96
• I guess it tends to positive infinity, so that was a responsible statement.--3-21-96
• If you don't understand what I'm going to say here, that's fine.--3-21-96
• Here is a model based technique with a bad model. Here is one with a better... less bad model.--3-21-96
• This seems like an exercise in futility here, but bare with me.--3-21-96
• You may say this is hopeless. Let's go home. But you can't go home. We need money to repair our homes, go on vacation, ...--3-21-96
• This is not the Generalized (Gaussian Markov Random Field). It is the (Generalized Gaussian) Markov Random Field. Generalized modifies Gaussian.--3-21-96
• You know, in the beginning of time, people could have decided not to put the one half in the Gaussian density.--3-21-96
• Typically the world has been nice in the past when we've tried to do estimation. Now...--3-26-96
• I could write this out, but I don't feel like it.--3-26-96
• Bite my tongue, this can be thought of as a fuzzy logic type problem.--3-26-96
• I could basically give you the answer right now, instead of going through the derivation because you can write the answer on the back of an envelope and it will always be right.... Having said that, of course, you won't REALLY understand it unless we go through it in all its gory details a few times.--3-26-96
• The following is strange but true. I am not going to attempt to give you any reason for what I am about to do. Just accept it.--3-26-96
• This is where the magic occurs. This is basically the EM algorithm. The rest is just pushing equations around.--3-26-96
• This is the absolutely essential inequality [Jensen's] that is responsible for all of the information theory... well almost.--3-26-96
• This is good because you could sit around for years and piddle around with this problem and not get the answer I am about to write down.--3-26-96
• Being a linguistically challenged person, I need all the help I can get with spelling and so on...--3-28-96
• I'm getting off on a tangent here, but there may be some of these words that you may want to remember.--3-28-96
• Okay, I'm going to write down formulas here, but they are not going to be right because I don't remember them.--3-28-96
• I'm running out of letters.--3-28-96
• Here's the moment you've all been waiting for... The next project.--4-2-96
• That's the problem with spell checkers. They don't tell you about words that are spelled correctly.--4-2-96
• Good thing this is not "fee" or this would be "fee fie".--4-2-96
• You must understand it if you noticed that [a mistake].--4-2-96
• Your project should be novel in that it has never been done before and that you have never done it either.--4-2-96
• Unfortunately, they decided to do that at the beginning of time and now we're stuck with it.--4-2-96
• If that doesn't make sense, you probably need to think about it a little while.--4-4-96
• I'm just using my standard abusive notation.--4-4-96
• I'm running out of blackboard space. I wish I had your handwriting, Chris. Then I could give you all little telescopes.--4-4-96
• [Mid-way through writing an equation on the board] Oh, did you see they caught the unibomber?--4-4-96
• You've heard of mathematical induction? You see one black bird, then another, then another. Therefore all birds are black.--4-4-96
• You're first assignment is to go back and check my equations. The article on this stuff has errors in the equations.--4-4-96
• This is going to sound ridiculous, but a good approximation of N large is N = 1.--4-4-96
• I want to give a warning about this, so pay attention for a little while. If you try to use this algorithm [the Metropolis Algorithm] without understanding it, you could waste months.--4-9-96
• Uh oh. I'm using N for two different things... well, something is going to have to change.--4-9-96
• I'm not saying that it is, and I'm not saying that it isn't.--4-9-96
• This seems natural, but this is another of those things with a gigantic warning sign on it.--4-9-96
• So basically this is the main thing you are supposed to learn from the last two and a half weeks. If you learn this, it will all have been worth it.--4-9-96
• The natural force to do it the wrong way is just too overwhelming.--4-9-96
• Don't ask me why I'm writing it this way. If you want to get the answer, you have to write it this way.--4-9-96
• I'm not here today so if you see me today, pretend I'm not here.--4-11-96
• Feel free to ask questions. The dumber they may seem to you the better they probably are.--4-11-96
• I define this to be f(p). Why? Because I don't know what it is, so I'll feel better if I give it a name.--4-11-96
• Let's say you had to sit down and write this program. What would you need to do?... What is the first letter you would type?--4-11-96
• If you generated an image from this Markov Random Field model and it looked like a sailboat, you'd be astounded. You may quit your job and go into missionary work.--4-11-96
• A mathematician may say, "This is impossible. Let's give up and go home." As an engineer, you don't have that option. Even if the problem is impossible, you still need to come up with an answer.--4-11-96
• I say these things because as an engineer it is important not to be arrogant about what you are doing.--4-11-96
• There are many papers that say problems are massively computationally intractable, but when you sit down and think about it calmly, it may not be that difficult.--4-11-96
• You take an animal or human and you inject something in them. If it's a highly lethal chemical, you tend not to use a human.--4-11-96
• I want to emphasize that you don't have much time for the projects.--4-16-96
• It's like taking off a band aid really quick. [An in class exam (instead of a takehome)]--4-16-96
• You can all go out and get intoxicated and forget about it [the exam].--4-16-96
• Okay, somebody write this down so I can get a copy.--4-16-96
• Is next week dead week? But you are all grad students... It's not an exam, it's just a high pressure situation. [Oral presentations]--4-16-96
• There was something else like that [misspelled word] that was bothering me earlier during the class, but I decided not to be juvenile.--4-16-96
• Oh no, now I'm suffering from permanent brain damage.--4-16-96
• The problem is that I have contaminated you with my way of looking at this problem. I think this is an easy problem, but the world at large thinks that it's horrendously difficult and have come up with many sick and twisted ways of solving this. If we had more time, I would like to go through these twisted ideas, but it would be much easier to just give you the answer to this impossible problem.--4-16-96
• I think the reason this had not been solved is that nobody ever bothered to write the equations down.--4-16-96
• There are no bad optimization methods, just bad problems.--4-16-96
• Some possible optimization methods:
  1. ...
  6. All the other methods that I am not familiar with.--4-16-96

  I know I usually keep you over a few minutes, but we're at a natural breaking point so I'll let you go.--4-16-96

  So is everyone's project done?... I tried to say that with a straight face.--4-23-96

  I'll try to get your midterms graded by Thurs. [The last day of class]--4-23-96

  You haven't really done anything practical in this class in terms of the homework.--4-23-96

  A picture is worth 1000 equations... or at least fifteen.--4-23-96

  Student: Why is that easier?

  Prof: It's a little faster. Another answer to that question is, "I have a subroutine that does it.--4-23-96

  This is the moment of truth. This is where great revelation occurs. If not, it's all been wasted.--4-23-96

  These matrices are ridiculous. We can't even store these on a computer. They're too big.--4-23-96

  Rather than try to derive it analytically, which takes way too long and probably wouldn't be too insightful, I'll show you a picture and that will convince you that this is true.--4-23-96

  Hold on, you let me write something here that is absolutely wrong.--4-23-96

  I didn't get your midterms graded. I'll do them tomorrow night.--4-25-96

  I'd like to keep teaching for the next three weeks, but we're out of time, so here's three weeks worth of reading material. [Hands out a bunch of papers.]--4-25-96

  This is the part where the magic occurs.--4-25-96

  The problem is that the only way these two pixels are related is through the president of the university and the state senate.--4-25-96

  Cars shouldn't be in the middle of the sky unless they are Chitti-Chitti-Bang-Bang or something like that.--4-25-96