The Gilt technology organization. We make gilt.com work.

A Brief Chat with NYC Machine Learning Organizer Paul Dix

image

On Thursday, January 16, Gilt VP Personalization Erik Lumer will be the featured presenter at the NYC Machine Learning group’s monthly meetup. Erik will talk about the art of personalization in flash ecommerce–covering the unique challenges for personalization posed by an online flash sale business, the methods that Gilt has tested to foster novelty in the recommendations of personally relevant products, and other related topics. RSVP here if you’d like to attend Erik’s talk, which starts at 7 PM at Pivotal Labs!

With nearly 4,500 members, NYC Machine Learning is one of the most popular technology-related meetup groups in New York City. What has been the key to its success? And how has the group changed over time? Recently I asked meetup co-organizer (and Errplane CEO, InfluxDB core committer, and Service Oriented Design with Ruby and Rails author) Paul Dix to share his thoughts on these and other questions:

Gilt: You started NYC Machine Learning in 2009. What sorts of meetups were you hosting then, and how were they different compared to your current events?

Paul Dix: In the very beginning I ran NYCML as a reading group. The format was something like this: I’d select a paper in machine learning research or we’d find one as a group, everyone would read the paper, and we’d meet once a month to discuss it. The first two months followed this format. In the third month, someone suggested a talk they wanted to give. Ever since then, that’s been the format of the group. Once a month, we get together to hear a talk from someone in academia, research, the start-up community, or business. The talks have covered a wide range of topics, including Artificial Intelligence, natural language processing, machine learning, search, information retrieval, big data infrastructure, and visualization.

Of the (76!) meetups you’ve hosted, which have been your favorites?

There have been so many fantastic talks that it’s hard to pick. Taking a quick look back, I’ll just pick four that caught my eye (of course, there are plenty more that I’d include in this list). In ascending order by date:

  • Princeton Professor Rob Schapire, “Playing Repeated Games: Theory, an Algorithm, Applications.” I remember covering AdaBoost in class, so I was really excited to have him come speak. And his talk was really entertaining.
  • CMU Professor Cosma Shalizi, “So, You Think You Have a Power Law, Do You? Well Isn’t That Special?” Cosma is a really well known statistician and has a great speaking style. This one was kind of a revelation to me as I had no idea that many things you think are power laws actually aren’t.
  • Microsoft Machine Learning Research Scientist John Langford’s “Terrascale Learning.” John had given a talk at NYCML before, but this one was particularly awesome. The approach he used here blew my mind and made me completely rethink how Hadoop and distributed computing work.
  • Peter Norvig. Max, my co-organizer, roped Peter into presenting, and we just put his name on the event. In my favorites because, come on, Peter Norvig!

Why did you start the group?

I started it for a couple of reasons. First was a purely selfish desire to try to stay current in machine learning and other topics that interest me. The second was that I wanted to help grow a machine learning community here in NYC. In 2007 I did an internship at Google, which holds internal machine learning meetups every two weeks. Google’s meetups weren’t open to the public, but the people who came and spoke at the meetings were great. [Google Director of Research] Peter Norvig, [Stanford professor] Mehran Sahami, and some other really heavy hitters would show up and participate in discussions or give talks. I wanted to see if I could reproduce the high caliber of those meetings for everyone here in NYC.

What do you think are the key elements of your group’s growth and success?

The most important factor has been consistency. Meeting up every month is huge. Every time I schedule a meetup, we get new members. The second factor: obviously fantastic speakers. Having presenters who come from academia and business has been great as well. With 4,500 members, we can’t get everyone in a single room–so with varied topics, there’s sure to be something every couple of months that any individual will probably be interested in. Finally, having a regular space has been really helpful. When everyone knows where to go, it’s easier. You hosted us for a while in the beginning, then we spent a few years at AOL, and for the last few years we’ve been at Pivotal. Having access to their space every month has been so helpful.

As a seasoned meetup organizer, what do you think are the essentials to organizing a great event?

All the obvious things. Have a space that’s easy for people to get to and find, offer a good setup for presenters, get an interesting speaker, and ask the speaker to provide as much detail as they can about the talk they’ll be delivering. I usually tell speakers to keep the talk between 30 minutes and an hour. People’s attention drops off after that, and besides, you want to give people time for questions and networking. I think it’s also a good idea to keep the Q&A time limited. People are there as much for the talk as they are to meet and talk to other people. Make sure there’s time built in for it.

Based on your observations, have any of the machine-learning topics that were “hot” in 2009 fizzled out or faded away?

I think recommender systems were much hotter in 2009 than they are today. I attribute that to the Netflix Prize, which was still going then. Once the team won, and Netflix decided not to implement the algorithm, it seemed everyone’s excitement calmed down. Recommendation systems are still important, but it seems they’re not as hot as they once were.

Other than your own group, what are some of your favorites in NYC?

The NYC.rb meetup is the one I’ve been going to the longest (since 2005). It’s a great group of people, and the topics reach well beyond just Ruby. I’m probably about 10x further along in programming and my career because of that group alone. Other ones of note: Ember.js, CTO School, the New York Hadoop User Group, and Go Language New York.

What are your goals for NYC Machine Learning in 2014?

I’d like to get someone from IBM’s Watson to come give a talk. I’d also like someone to give a talk on deep learning. Other than that, I’ll be happy as long as we meet up at least once a month and have a great speaker to learn from.

Thanks for chatting, Paul!

Erik Lumer 7 personalization 12 NYC Machine Learning 1 Gilt 340 Gilttech 18 Paul Dix 2 meetups 35 machine learning 8 ecommerce 35 recommendations 3 big data 17 data visualization 5 information retrieval 1 AI 2 artificial intelligence 2 natural language processing 1 Netflix Prize 1 Peter Norvig 1 Mehran Sahami 1 Google 4 NYC.rb 1 CTO School 2 New York Hadoop User Group 1 Go Language New York 1 Ember.js 2