Past Meetup

Deep Dive into TensorFlow #3

This Meetup is past


1000 1st Avenue South #500, Seattle, WA 98134 · Seattle, WA

How to find us


Location image of event venue



Please register HERE (

Many thanks to PayScale ( for hosting and sponsoring the Tensorflow meetup!

Details: Please enter on Occidental side of building, which is the side that faces Centurylink stadium. (Google maps will point you towards 1st ave side of building)

Parking Information: century link parking garage ($8) or the North Lot next to Century link ($13).


6:30 - Doors open. Networking. Pizza and beer

7:00 - Software Patterns in TensorFlow by Garrett Smith

7:45 - Q&A break

7:50 - Extracting information from text with recurrent neural networks and Keras by Mike Stark

8:30 - Q&A break

8:40 - Wrap-up.


Software Patterns in TensorFlow

TensorFlow is a flexible, general purpose computational library that's used to implement a wide range of machine learning models. Its flexibility however presents a challenge: how do teams discover and apply effective software patterns in their projects? In this presentation, Garrett Smith, founder of Guild AI, will share his experience working with dozens of TensorFlow projects and discuss patterns that work well and those that don't when writing TensorFlow code.

Garrett will cover:

- Project structure
- Variable naming conventions
- Canonical functions and workflow
- Parameterization using flags
- Logging and retraining experiment results
- Conventions for serving trained models
- Lessons from TFLearn and Keras

Presenter: Garrett Smith, founder of Guild AI

Garrett Smith is founder of Guild AI, an open source toolkit that helps developers gain insight into their TensorFlow experiments. Garrett has over twenty years of software development experience and has managed teams across a wide range of product development efforts. His has expertise in building reliable, districuted back-end systems and in operations. Prior to founding Guild AI, Garrett led CloudBees PaaS division, which hosted hundreds of thousands of Java applications at scale. Garrett is a frequent instructor and speaker at software conferences and an active member of the Erlang community, maintaining several prominent open source projects.


Extracting information from text with recurrent neural networks and Keras

The Concur application ExpenseIt allows travelers to take pictures of their receipts and automatically create expense reports from them. The ability to create expense reports automatically relies on optical character recognition (OCR) technology and machine learning to extract important information from the OCR text. Important pieces of information we want to extract from the receipts include amount, date, currency, vendor and location. We use a variety of machine learning techniques to extract this information from OCR text which often includes flaws introduced by the OCR process. We have begun to rely, increasingly, on recurrent neural network sequence-to-sequence models for this work. I will present a neural network design involving a one-shot version of attention that is remarkably successful in extracting this information. The model and attention mechanism are coded straightforwardly in Keras which can use TensorFlow as a backend.

Presenter: Mike Stark, Data Scientist at Concur

Mike Stark was an academic astronomer for many years concentrating on black holes and neutron stars observed via satellites. In 2015, he followed his growing interest in machine learning out of academia and into Concur. At Concur, as part of a data science group, he is working on machine learning solutions to various problems created by and/or addressable with large volumes of data. He is particularly interested in the surprising power of recurrent neural networks.

Mike earned his Ph.D. in physics at the University of Maryland, College Park.

Please register HERE (

Attendees (1)

Go to Attendee List