Profile

Cover photo
Matt Gardner
Works at Carnegie Mellon University
Attends Carnegie Mellon University
268 followers|572,442 views
AboutPostsPhotosVideos

Stream

Matt Gardner

Shared publicly  - 
 
I'm at a thesis proposal at CMU, and I'm not totally sure what to think of this racial disparity:

Number of white male students in the room: 2 out of about 45 (the rest are almost all Asian (including Indian), maybe 25% female)
Number of white male thesis committee members: 4 out of 4

Does this mean that minority students don't have hope of getting tenure, or that white male students don't have much hope of getting into CMU, or that the racial distribution of tenured faculty is in for a big change?
2
Andrew McNabb's profile photoJordan Boyd-Graber's profile photoMatt Gardner's profile photo
3 comments
 
An LTI faculty member told me that they just hired a marketing and outreach person, with the goal of getting more American grad students, which I thought was kind of funny. When I started I was one of maybe 5 or 6 out of 60 or so. Partly this is just a CMU demographics phenomenon, but, as Jordan points out, it's also a broader issue. Also in CMU's defense, there are a number of non-white-male faculty members, they just didn't happen to be on that particular committee.
Add a comment...

Matt Gardner

Shared publicly  - 
 
PhD thesis successfully proposed.  Now to actually finish the work...

If you're interested in seeing the proposal document: http://cs.cmu.edu/~mg1/proposal.pdf
12
Daniel Zappala's profile photoMatt Gardner's profile photoNeil Self's profile photo
7 comments
 
Thanks, yeah, Kelli started a master's program at San Jose State and I got a job at the big G.
Add a comment...

Matt Gardner

Shared publicly  - 
 
A really nice paper on tensor factorization models for KB inference. This paper had a big impact on my current research direction.
I really enjoyed thispaper,and it had a significant impact on my research direction. I am looking at combining compositionalmodels of KB inference with mode...
1
Yuanfei Luo's profile photoMatt Gardner's profile photo
2 comments
 
I'm hoping to get the work mentioned here ready for submission to EMNLP, but I've got a number of other papers I'm also working on.  Either way, it should be published somewhere within a couple of months.
Add a comment...

Matt Gardner

Shared publicly  - 
 
And lastly (on the topic of skip-gram, anyway) are three papers that were reactions or follow-ons to the skip-gram idea.  One says that "predict" models (like skip-gram) are definitely better than "count" models; the other two show "count" models that beat skip-gram.
Finishing up (for now) my reading about skip-gram, I'll summarize three papers that providedifferent reactions or follow-ons to the skip-gram papers, all...
1
Add a comment...

Matt Gardner

Shared publicly  - 
 
The second skip-gram paper.

(Can anyone tell me where the name "skip-gram" came from, by the way?)
The second word embeddings paper I'll discuss is the second main skip-grampaper,a follow on to the original ICLR paper that basically drops the CBOW mode...
1
Add a comment...

Matt Gardner

Shared publicly  - 
 
"Moral elevation" - I haven't heard the term before, but I know the feeling.
I was more polite when I carried the briefcase the Mormons gave me.
1
Add a comment...

Matt Gardner

Shared publicly  - 
 
I've read and thought a lot recently about net neutrality and competition issues with ISPs, and what the proper policy response should be.  This is a great article describing what now seems to me to be the best solution available.  Basically, we should force those who own last-mile infrastructure (which is and will always be a natural monopoly) to lease access to that infrastructure on a wholesale basis.  This is successfully done in many other parts of the world, and in the US with other technologies.  We should do the same for internet access.
5
Peter Aldous's profile photoDaniel Oliphant's profile photoMatt Gardner's profile photo
8 comments
 
And that's a question I don't have a good answer to. How do you influence the political process to get the FCC to do what it should? I have no idea.
Add a comment...
Have him in circles
268 people
Corina Negrea's profile photo
Carlos Kemeny's profile photo
Frederick Kintanar's profile photo
Ni Lao's profile photo
Helen Hall's profile photo
Pricing Healthcare's profile photo
Teruko Mitamura's profile photo
Mark Schofield's profile photo
Nicolas Maillot's profile photo

Matt Gardner

Shared publicly  - 
 
What should you do when your scientists come to conclusions that you find politically unsavory? Listen to their evidence? Oh, heavens no, you shouldn't do that. Just insult them and defund them. Obviously.
Appears to be part of a concerted attack on climate research and responses.
5
1
James Carroll's profile photo
Add a comment...

Matt Gardner

Shared publicly  - 
 
Reading some more papers on word embeddings. One paper uses dependency edges instead of words as contexts for skip-gram; one compares a neural translation model to skip-gram and related models; and one looks more closely at the objective optimized by skip-gram.

Skip-gram is nice because it is really simple and thus scales really well. And you can give it different information in the form of different contexts, and it will learn different things. But has anyone tried to learn a deeper representation of language, similar to the deep representations now commonly used for images and speech?  A quick search didn't bring up anything particularly relevant.
Ok, one last round of word embeddings papers, then I'm on to research that's a lot more relevantto my current work. Here I'll look at three pape...
1
1
Kirill Brylev's profile photo
Add a comment...

Matt Gardner

Shared publicly  - 
 
I've needed to learn more about tensor math and tensor decomposition for things I'm currently working on.  This is a good overview.
2
Add a comment...

Matt Gardner

Shared publicly  - 
 
Well, I've decided (once again) that I need to be spending more time reading research papers, and I still think that writing about them is a good way for me to distill my thoughts.  I set up a simple blog on github pages where I can keep these formatted a little better than I used to by just putting them on Google+.  I'll still share them here, in case anyone finds them useful.

My first topic to read about was the recent craze over skip-gram and related word embeddings papers.  This is the paper that originally introduced the skip-gram model.
I'm a bit late to the word embeddings party, but I just read a series of papers related to theskip-gram model proposed in 2013 by Mikolov and others at G...
1
Bryan Perozzi's profile photoMatt Gardner's profile photo
2 comments
 
Thanks for the pointer.  That's on my list as the next paper to read.
Add a comment...

Matt Gardner

Shared publicly  - 
 
Fascinating.  And this seems like a great example of a very comprehensive science experiment - there was a lot of work crammed into that one paper.
Saccharin and other sweeteners alter the bacteria living in our guts.
2
Add a comment...
People
Have him in circles
268 people
Corina Negrea's profile photo
Carlos Kemeny's profile photo
Frederick Kintanar's profile photo
Ni Lao's profile photo
Helen Hall's profile photo
Pricing Healthcare's profile photo
Teruko Mitamura's profile photo
Mark Schofield's profile photo
Nicolas Maillot's profile photo
Education
  • Carnegie Mellon University
    Computer Science (Language Technologies), 2011 - present
  • Brigham Young University
    Computer Science, 2004 - 2011
Links
Work
Employment
  • Carnegie Mellon University
    Phd Student, 2011 - present
  • Google
    Research Intern, 2014
  • Google
    Software Engineering Intern, 2013
  • Google
    Software Engineering Intern, 2011
  • Google
    Software Engineering Intern, 2009
  • Allen Institute for Artificial Intelligence
    Intern, 2014
Basic Information
Gender
Male