TWIML Meetup Planning

150 150 The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)

Hi Everyone!

*** UPDATE: Visit for information on the current meetup! ***

Thanks for your interest in the paper reading group. I’m excited that so many folks have expressed interest. I’d like to use this page to share my thoughts on how this will be organized, the input I need from the community, and how you can help. Please submit your thoughts via the comments.

  • Event: This will be run as a monthly online meetup. We could do the first one as soon as mid-August, giving participants a full month to read the first paper. We would continue on a monthly basis, and announce the next paper at the end of each meetup and by email.
  • Presentations: The meetup will consist of a mix of community-driven presentations (75-90%) and invited presentations (10-25%). By this I mean most of the time members of this community present papers of their choice at the meetup–not necessarily their own. Occasionally we will bring in researchers to present their own papers, which I think would be fun twist.
  • Meeting Format: The meetups will consist of a 45-60 minute presentation of a given paper, plus a Q&A period that goes “until,” i.e. until the last participant drops out of the meetup. I’d expect them to go two hours on average. The presenters are expected to prepare and deliver slides to support their presentations.
  • Technical Level & Presentation Goals: This will vary from week to week, and depend on the presenter and the community. I expect it to be fairly technical, but I think the job of the presenter is to:
    • Explain the broader context of a paper. Why is it interesting and significant? What assumptions does it make?
    • Take us beyond the equations. What’s the point? What do the equations mean? Where do they come from?
    • Show us; not just tell us. This might be too much to ask, but I think it’d be great if every presenter at least tried to implement the paper and did a demo, if appropriate to the paper.

    Beyond these points, I think folks can get more technical in the Q&A.

  • Days & Times: I’m not sure if folks see this as an extra-curricular activity that should happen during the weekend, or a career/study activity that should happen during regular hours. One thing I’ve seen well is to stagger the hours from meetup to meetup and around the presenter’s availability so at least no one is perpetually excluded or inconvenienced.

Papers to Consider:

  • Pedro Domingos (2012). A few useful things to know about machine learning. [High level]
  • Classic DL Papers: LeNet, AlexNet, GoogLeNet, ResNet
  • GANs (2014)
  • Generating Image Descriptions (2014)
  • Long Short Term Memory (LSTM) (1997)

That’s about all I have right now. Please let me know what you think and any ideas you have. Also let us know where you’re located geographically, so we have a sense of the time zones that may work for you.

I’d like to identify 3 or 4 community members to help with the organization of this. Please let me know if you’re interested.

  • Duncan

    This looks like an excellent structure overall! Monthly meetings sound like a good frequency and having presenters also share the motivation/application of the methods presented in addition to some nitty gritty sounds like a good format to keep everyone in the loop.

    The one thing i’d possibly change is the number and length of the presentations. Depending on the number of people present 2 hours per presentation could get lengthy. Presentation length should probably vary with experience e.g. guest researchers or wrote the paper in question 2+ hours down to 30ish minutes for those just reading the paper but not involved in its publication. This would ofcourse depend on how many people present as well.

    I would be interested in helping organize and currently am in the Boston/Cambridge MA area.

    Looking forward to hearing more about this!


    • Joshua Manela

      Yea, I would agree with Duncan. I love the format but 2 hours as a gold standard seems to be a bit lengthy. As a side note, would we be recording our discussions and publish them somewhere for other people to listen to later? I think that would be a really cool for people who want to keep up to date with current research in the community but may need some time to take in all the material.

  • Nikola Kucerova

    Hi Sam!

    I would love to help with organizing this as well. Discussions sound like a great idea and I also agree that 2 hours might be a bit too much. I am currently based in Sydney and I really don’t expect the days and times to suit me so I’m glad that there will be recordings on YouTube. Also, I would like to suggest another paper – One Model To Learn Them All (2017)



  • Domenick Poster

    I’m a PhD Student in Deep Learning, Computer Vision, and Biometrics. I have a habit of spending too much time coding and not enough time reading papers. I would be more interested in reading if I could discuss and hear others comments about the paper. I am in EST timezone.

  • Jan Zyśko

    Hello! I would like to participate as well. I’m a Data Scientist and starting an ML PhD in Poland. 2h would be fine for me, as I would want it to be very technical. With <1h I think it would be just fluff.

  • Roslynn Ricard

    Hi, this reply is a bit delayed but I am also new to the world of machine learning but work as a Controls Engineer in the Bay Area. So applying ML to Industrial Controls / Robotics / Manufacturing is certainly of interest. I’m a big fan of the podcast and think this additional initiative is a great idea.

    • sam

      Thanks, Roslynn. Glad to hear you’re enjoying the series. I’d love to learn a bit more about your work and what you’re up to there. Let me know if you’d like to chat sometime.

Leave a Reply

Your email address will not be published.