Vergangenes Meetup

Coding session - learning by doing: Keep-Current Project

Dieses Meetup liegt in der Vergangenheit

52 Personen haben teilgenommen

WeAreDevelopers Office

Doblhofgasse 9 Tür 14 · Wien

Wie du uns findest

Pay attention: use the Doblhoffgasse 9 entrance, with the sign: WeAreDevelopers - Reception. There's a buzzer on the right side of the main door. We're on the 4th floor, door 14. The meeting room is on the left.

Bild des Veranstaltungsortes

Details

In this meetup sessions, we learn coding (python) and machine learning, by doing it!

This evening, we will host Jorge M. Loaiciga, who will guide us how to deploy TensorFlow models in production.

Abstract
------------
In this talk we will point out some of the common problems that arise when deploying machine learning models for serving. We then describe what the open source project known as `Tensorflow-Serving` offers for tackling said issues and mention some of its more advanced features. Additionally we list the components of the underlying technology and finalize with a brief walk-through / demo on how to both deploy an object detection model (probably using the Keras implementation of YOLOv3) and request service from the client side.

Outline
----------
1. Usual complications that arise when deploying ML-based services:
- Throwing code over the metaphorical wall (from development to operation)
- Ease of development and maintenance, portability (environment and hardware), reliability, scalability (serialized graph execution, gRPC, batch processing)
- Bonus ~Meme~: What advanced features exist? (online training, model versioning)
2. TensorFlow serving's approach (containerized c++ server, reachable through gRPC and more recently, through REST)
3. A glance at the the underlying tech (gRPC, Docker, probuffer and serialized model graphs)
4. Demo: Exporting a Keras model for deployment & basic building blocks required on the client side for interacting with the remote service

----

Jorge M. Loaiciga received a MSc. degree in Electrical Engineering from the Korean Institute of Science and Technology in early 2018. He is a developer for COZYO (interior design automation and AR visualization) since mid 2018. His research interests include computer vision and techniques for reduction of DNN (Deep Neural Network) computation requirements.

---

What is Keep-Current?

Keep-Current is an educational open source project that filters machine-learning related academic papers, to help you find the most suitable content for you.

Through building and contributing to the project, one can gain experience in different fields:
Software Architecture and design; micro-services and DevOps; data engineering; Data science and analytics; Machine Learning and Natural Language Processing; Python programming experience; and much more.

The purpose of this meetup is to help you get started with contributing to the project.

Please visit the Keep-Current project on github to learn more about its architecture (https://github.com/Keep-Current).