Distributed Multi-Device Execution of TensorFlow for IoT

by Sophia TurolMarch 21, 2016

Learn about TensorBoard’s capabilities in visualizing learning, the challenges of using TensorFlow for embedded systems, and multi-dimensional IoT data.

Below are the videos/slides from the TensorFlow Munich meetup—sponsored and organized by Altoros on March 1, 2016.

Table of Contents

TensorFlow overview

In his session, Ankit Bahuguna of Cliqz provided an insight into TensorFlow usage, the library’s mechanics, and implementation. To demonstrate TensorBoard’s capabilities in visualizing learning, he exemplified two demos on linear regression and visualizing word embeddings in TensorFlow. Finally, he shared the results of the recent benchmarks and outlined the likely scenarios of evolution.

Below, you can check out the full slides by Ankit.

Distributed multi-device execution of TensorFlow

Sebnem Rusitschka, Senior Key Expert for Cyber-physical Systems at Siemens, provided an overview of TensorFlow from a distributed computing perspective. She also elaborated on the perculiarities of embedded systems when using TensorFlow and highlighted the associated challenges. Finally, Sebnem talked about multi-dimensional IoT data, tensor networks, and arrays/indexing.

Fireside chat

Later on, Alex Osterloh of Google and Artyom Topchyan of Reply discussed:

What kind of tasks TensorFlow is built for?
The industries / projects TensorFlow is applied within and the feedback received
Recommendations for those getting started with TensorFlow
The biggest challenges TensorFlow has at the moment
Contributions of the community that can be of help
The impact of open-sourcing TensorFlow

Join our group to stay updated with information on the upcoming meetups.

About the speakers

Ankit Bahuguna is Software Engineer at Cliqz, where he develops deep learning solutions to tackle the problem of low latency in web search. Ankit’s current work is focused on query embeddings, where the semantics of a user query is preserved in a fixed dimensional mathematical vector—trained using deep neural networks.

Sebnem Rusitschka has 10 years of experience in translating Internet-scale innovations to the world of industrial machines and processes. The area of her expertise includes digitized automation of infrastructures, especially, electrification. In 2015, she became Senior Key Expert for Cyber-Physical Systems at Siemens with a mission to bring together experts from various disciplines of machine learning, control theory, and distributed computing to achieve adaptive automation systems.

Alex Osterloh is Big Data Solutions Engineer at Google. He graduated from Ludwig Maximilian University of Munich with a master’s degree in computer science and had experience as Presales Engineer at Hewlett-Packard before joining Google in 2010. Alex’s current job involves converting big data to smart data and leveraging Google Cloud Platform for doing large-scale data crunching.

Artyom Topchyan is Technical Lead at Data Reply. He graduated from Technical University of Munich with a master’s degree in robotics. As Research Assistant, Artyom managed to integrate a number of machine learning–based methods into the university’s computer vision library. He also has experience in natural language processing. Artyom developed a system for large-scale extraction and analysis of job descriptions, helping to analyze the required competencies and competency dynamics in the job market.