Portland including Mount Hood and Oregon Convention Center Flowers

The TensorFlow Community Day (a first of a kind – #TensorFlowDay) event took place on Tuesday July 17, 2018 in beautiful Portland Oregon at the OSCON 2018 conference from O’Reilly. There were TensorFlow related talks intended to inform and help build community around Tensorflow. There were plenty of opportunities to have technical discussions and to network. All the talks were recorded and will be available on Safari. The event was sponsored by Google & IBM.

The community day talks were accompanied by a TensorFlow hacking room which had 8 sets of hacking table leaders. People joined tables of their choice to start hacking or to have in-depth discussions or to learn – and some teams grew to over two tables, at times, e.g., Kubeflow and video analytics tables. Some leaders set up slack channels to keep the community going that they created during the day, and some leaders had people join existing slack channels. All leaders got feedback and new ideas. There were wifi problems but there were wired connections too (most people needed HDMi to ethernet adapters to connect to the cables – and the adapters are different for Macs). Eventually an additional wifi hub was placed in the hacking room, resolving the issues (Thank you Marcus Chang & O’Reilly).

About 180 people attended the day overall and people moved between the talks and the hacking room. The hacking team leaders presented their projects briefly in the morning, to attract hacking participants, and then provided a readout at the end of the day including potential follow-ups – which was positive.

In summary, the TensorFlow hacking room worked very well alongside the room with TensorFlow talks, with people moving between the two rooms depending on their interests & what was happening in each room at the time. The fact that the hacking leaders were from different institutions and backgrounds, and were all enthusiastic, made for a wonderful experience. You can see more about the hacking tables if you scroll down. I hope we have similar hacking and on-boarding rooms at the upcoming conferences such as Think 2019. JupyterCon in NYC run a similar community sprint day and I am looking forward to it, having participated last year. Thank you Edd Wilder-James, Marcus Chang, Nancy Mohamed, the Google team, the IBM Team, the hacking leaders and participants, for the great support and stunning TensorFlow Day.

The TensorFlow Build Hacking Table – From https://twitter.com/sumalaika/status/1019347833114095617

The Hacking Tables

These were the TensorFlow hacking tables, photos and some of the follow-ups:

Related Materials:

Join us in the hacking room at the TensorFlow Community Day https://conferences.oreilly.com/oscon/oscon-or/public/schedule/detail/70899 on July 17 in Portland, where we will have about 10 tables, of up to 10 people working on projects, or sharing technology skills and ideas. Each table will have, a flip chart or white board, a power strip & one or two human leaders.
If you are a hacking room leader, you will:

  • Present (briefly) to the attendees of the TensorFlow Community Day what the activity is at your hacking table – and the skills of the people you would like to join your table – and the goals for the day
  • Lead and work with your volunteer team (up to 10 people) through the day until 4pm
  • Present (briefly) the achievements of your hackers in the main tent

If you are a hacking room participant, you will:

  • Join a team and work with them as much as a possible through the day – Your activities could be coding, testing, documenting, creating a demo, brainstorming etc. Bring your laptops. You will have the opportunity to network with a wonderful set of people and build long-term technology relationships.

If you’re already registered for OSCON 2018, all you need to do is turn up. If you’re not registered, you can sign up for OSCON’s Expo Plus Pass on their registration page, and TensorFlow Day will be open to you. Refreshments and lunch will be provided for all registered participants.


Hacking tables and Participant Prerequisites:

  • R & TensorFlow – led by: Gabriela de Queroz & Augustina Ragwitz
    —– Getting started with Tensorflow and R
    —– Identify a interesting dataset to be used in the demo (from a predefined collection)
    —– Create a end-to-end workflow: from pre-processing the data to final model
    —– Demonstrate features we call out in our talk — actual live demos
    —– Ideal participant prerequisites Should have a reasonable amount of R experience and basic knowledge of machine learning methods and algorithms.

  • Pokedex – Identify Pokemons with Google AIY Vision Kit – led by: Alex Kari, David Molina, & Al Kari
    —- Explore the basics of Google Collaboratory, a free, GPU enabled notebook environment
    —- Learn how to download an image dataset to train a neural network
    —- Discover how easy it is to retrain a TensorFlow ImageNet model with custom objects (Pokemons)
    —- Compile and deploy your retrained model to run on the Google AIY Vision Kit (RaspberryPI)
    —- Use your model to accurately identify new Pokemons in real-time
    —- Ideal Participant prerequisites: Reasonable knowledge of Python and basic understanding of machine learning concepts. Should pre-register at https://colab.research.google.com

  • TensorFlow Build – led by GĂĽnhan GĂĽlsoy, Yun Peng, Thomas Truong & Ted Chang
    —- A meeting of the TensorFlow SIG Build group: newcomers welcome!
    —- Bazel office hours, expert help with mastering Bazel.
    —- Review current issues with the TensorFlow Bazel build, collaborate on fixes
    —- Identify requirements for the Bazel team’s Q3/Q4 feature planning
    —- Set up a community custom build, using IBM Power as an example
    —- Ideal Participant prerequisites: Reasonable knowledge of any build process. Bazel skills is a plus but not required

  • Maths Kernel Library (MKL) & TensorFlow – led by Clayne Robinson & Guozhong (GZ) Zhuang
    —- help people get started with building TensorFlow with MKL support and using existing public whls and containers
    —- get a closer look the issues that off-the-street users face when they are using TF+MKL out of the box
    —- brainstorm as to how we can better include community TF+MKL contributions that don’t come from Google and Intel.
    —- Help the OS community understand how MKL optimizations have been implemented.
    —- Ideal participant prerequisite: Access to a Linux system (remote or laptop, VM or bare-metal) with either docker or gcc installed; decent grasp of C++ if you want to get into the guts of TensorFlow + Intel® MKL-DNN.

  • Distributed Video Analytics using Tensorflow Models – led by Vinay Rao & Santi Swaroop Adavani
    —- Video traffic will be 82% of total Internet traffic by 2021, according to Cisco. Video content analysis, capability to analyze spatial and temporal events in the video, is increasingly important in various domains including but not limited to health-care, entertainment, transport and home automation. At this hacking table, we
    ——- Help get people started with OpenCV and Tensorflow
    ——- Work with videos, understand the value of video data
    ——- Help the community to understand how to use tensorflow object detection models using OpenCV on videos
    ——- Identify strengths and limitations of current object detection models for real life applications
    ——- Brainstorm ideas on how to use video analytics data for improving quality of life
    —- Ideal participant prerequisites Python, OpenCV, and Tensorflow installed on their devices.

  • Kubeflow & TensorFlow – led by Michal Jastrzebski & Ankush Agarwal
    —– We will learn about setting up Kubernetes for machine learning
    —– We will use Kubeflow as base for our training
    —– We will learn how to write Tensorflow model that can be distributed with help of Kubeflow
    —– Together we will tackle real problem (Kaggle challenge) from start to finish and run it on cloud
    —- Ideal Participant Prerequisites Required knowledge: Tensorflow, basic Kubernetes, general machine learning ; Other requirements: account on Kaggle https://www.kaggle.com/

  • TensorFlow.js Demos and More – led by Sandeep Gupta, Ton Ngo, & Yi-Hong Wang
    —– Show TensorFlow.js demos
    —– Show TensorFlow.js code
    —– Identify areas for further development in TensorFlow.js
    —- Ideal participant prerequisites Have some knowledge of JavaScript, Node.js, & TensorFlow – Optional : Have Tensorflow installed on their devices

  • Call for Code with TensorFlow – led by Paul van Eck
    —– Learn about https://callforcode.org/
    —– Look at TensorFlow examples
    ———https://developer.ibm.com/code/patterns/create-mobile-handwritten-hangul-translation-app/
    ———https://developer.ibm.com/code/patterns/classify-art-using-tensorflow-model/
    —– Formulate a submission
    —– Start working on submission
    —- Ideal participant prerequisites Have some knowledge of TensorFlow. Register for cloud account https://ibm.biz/BdYRNi
Portland from Wikipedia – This file is licensed under the Creative Commons Attribution-Share Alike 4.0 International license.

Join The Discussion

Your email address will not be published. Required fields are marked *