About Me
Hey! I'm Matt Jones. I like making things do what they're not supposed to.
I currently lead the anti-spam engineering team at WhatsApp, focusing on scalable ways to detect and prevent abuse for 1 billion users.
Prior to that, I spent seven years at Facebook working on anti-abuse systems, from reverse-engineering malware and building high-throughput
classification systems to defending new products and leading teams in California and London.
Along the way, I protected against malicious ISPs, compromised telephone carriers, and numerous spammers.
Before Facebook, I earned my BS in Computer Science in 2009 from Stanford University.
I am interested in
building and breaking secure systems, understanding and visualizing large datasets,
and sometimes talking about elegant user design.
Otherwise, I enjoy cycling, skiing, sailing, and hiking.
For more blather about me, you can check out
my LinkedIn
Projects
I've worked on a number of projects for fun, class, or
work. Here's some info about a few of them:
Facebook's Linkshim
(2009-2012): A way of keeping Facebook users safe from
a number of different attacks, including from malicious URLs and
sites trying to track via referrer. See another note about how the
referrer-protection
part in particular works.
Faster than SIFT Object Detection
(pdf) (2009): A project
I worked on with Borja Peleato, looking at ways to improve
the effective but slow Scale-Invariant Feature Transform.
We used generic trees, built on top of random tree
classifiers, based on work by
Calonder et al (pdf) to search through video for instances
of a query object. We were able to improve SIFT speed by
5 times, at the expense of a higher false positive rate.
We hope to open-source our implementation, and will
post a link here when it's cleaned up a bit and ready
for general consumption :-).
Internet
Comment Classifier (pdf) (2008): A paper I
worked on with Prasanna Vasudevan and Eric Ma,
exploring ways to apply machine learning techniques
to classifying Slashdot comments. Perhaps our most
interesting finding was that we could classify
comments as good or bad most accurately using only
information about how many times the given user had
posted before - discarding all data about the actual
comment content. I worked on scraping 700k+
comments, calculating metafeatures, running Naïve
Bayes with different features and class types, and
visualizing the results. We'd love to hear any
comments you have, so feel free to email mkjones
< at > cs < dot > stanford
. edu.
Tournaments
Facebook App (2007): A facebook
application written from scratch in php for
Stanford's CS377W
class with
Blake Cutler and Brent Pirruccello (who's now
working
on college
admission statistics). I did most of
the database, object-relational mapping, utility,
security, and otherwise backend code, along with a
few UI elements. The application lets users set
up, administer, and participate in Tournaments for
any sport. It was created before and independent
of my employment at Facebook.
- Written from scratch in php
- Totally object-oriented design
- Link tracking for metrics and UI analysis to augment Google Analytics
- Filesystem caching of objects for performance
- Form IDs on all actions for eliminating double posts and mitigating XSRF attacks
- Access-controlled invitations
- AJAX for wall posts and other actions
SendRoses
Facebook App (2008): Another facebook
application I worked on with Blake Cutler and Brent
Pirruccello, written using the same framework I
originally developed for Tournaments, above. This,
too, was made before and independent of my
employment at Facebook.
Framework for Social
Web Applications (pdf) (2007): A paper I
wrote alone, also for CS377W, discussing what
makes a high-quality social application that
benefits all parties involved.
Photography
for fun and for class. I took intro to photo
and had a lot of fun shooting and
developing on black and white film using my dad's
old Canon
T90 ("The Tank"). However, shooting film is
O(n) time and cost for the number of pictures you
take. In search of a more cost-effective strategy,
I bought a digital SLR, which is O(1) cost but still
O(n) time (though with a much lower
constant). Check out
my flickr
photos!
Intel Paper
Semifinalist (pdf) (2004): I did research
on the behavior of bronze powder when vertically
vibrated at relatively high frequencies, including
the appearance of and cause for certain patterns.
The resulting paper was a Semifinalist in the
Intel Science Talent Search. More info, including
my friends' projects, at my Intel
page.
Intel
Papers
The following are some papers written by friends of
mine in
the Chem-Phys program at Evanston Township High
School and submitted to Intel STS /
Siemens-Westinghouse / other outlets for high
schoolers doing original research. If you're
interested in the program, reading more papers, or
maybe doing one yourself, check out Mark
Vondracek's
high school research website at ETHS.
If you want me to put yours up here, just shoot me
an e-mail, mkjones at cs dot stanford dot edu.
Intel Papers Some of the
same students that do research through the Chem/Phys
program at ETHS also made
a sick
video about the program, presumably for a senior
project. Check it out.