Tuesdays with Mimi
12 February 2013 by Daniel Lee
Unedited transcript of conversation. Contents are presented in the order they were spoken during the conversation.
On Mimi's Fourier Transform Skills
D: Hi
M: You made your own lunch this time. It’s absolutely vegan.
D: Yea, it’s almond butter with stuff. Like a bagel pizza. Any good new, interesting story?
M: Today, I forgot why, but I was browsing the web looking for something, and I found a blog post about me. I don’t know who it is...some student..
D: In what sense?
M: He was just remembering my 301 course, and he was commenting on my Fourier transform skills and how I can Fourier transform my way out of a paper bag. Not sure what that means.
D: You can Fourier transform your way out of... a paper bag?
M: Yea, a paper bag.
D: Sounds like a compliment.
M: Is it? I’m not sure.
D: Yea, like you are hatcheting yourself out of a paper bag.
…
On Machine learning, and related topics
D: Is that the only course you’ve taught?
M: Four different courses.
D: Four? I’m talking about undergrad courses...
M: Three different courses. I teach a graduate course on spring of even years.
D: What is it?
M: Decision Theory
D: Sounds like math.
M: It can be math, but I don’t teach it as a math course. I teach it as “okay, here is a bunch of techniques - I am going to teach you the common denominator of all techniques, how they are related, and how they are all trying to do the same thing.. They may not look like it, but they do more or less the same . Once they learn the techniques, I ask them to try out the techniques and get a feel for which techniques are good in what situations, and which they are not so good for.
D: Is that machine learning?
M: Machine learning is... no... The goal of machine learning is to write an algorithm to make decisions or thoughts or to classify cases automatically. If you know how things work really well, you can actually make an algorithm that does it - make rules with. That’s not machine learning. Machine learning is when you don’t want to understand the problem, you don’t want to make rules. You just want to learn the rules automatically from data. The machine learns, not you.
D: That does sound different.
M: Its at a border between mathematics and disappointment. And nowadays, there is a , I don’t want to say new branch of mathematics, but bunch of mathematicians been looking at this persistent homogoly, have you heard to term?
D: No
M: Persistent homolgy, you can pull the check sequence starting from data and then you do this iteratively using smaller and smaller epsilon, and if things persists, if the tend to follow a pattern you declare that you’ve understood.
D: Declare you’ve understood?
M: The homology of the pattern. It’s funny. It’s got this fancy names and concepts, but I think what it does numerically at the moment is very very basic machine learning. The oldest methods of machine learning.
D: But its framed in math sense?
M: Yes. And it has opened the door to contributions from mathematicians. The numerics may be behind, but hopefully it will change.
…
On my bagel pizza
M: You should take half of this.
D: No. It’s really good, and you will probably want more of it. It’s actually my roommates creation.
M: It’s his recipe.
D: Yes, his recipe. I always saw him eat it, and it looks really nice and smells really nice too.
M: That’s peanut butter, right?
D: No, it’s almond butter. Bit sweeter than peanut butter. I only tried cream cheese with bagels before, but I tried peanut butter spread and it was good. I tried almond butter, it was better. You’re not allergic to nuts are you?
M: No I’m not allergic to.. well no food but lots of other stuff. Like dust.
D: Good.
M: When a guy your age makes something to eat, we always say... how do I translate this.. a good party?
D: Good party?
M: A potential husband?
D: Ha. What's the word? It’s French right?
M: You can’t directly translate it, but that’s what it means.
D: But how do you say it? Maybe it sounds cool
M: “un bon à marier”
D: Ah.. can’t pronounce that. French is harder than I thought it would be.
---
