Presentation: TensorFlow: A Flexible, Scalable & Portable System


2:55pm - 3:45pm



Key Takeaways

  • Hear key design considerations of TensorFlow from the Engineering Director leading development.
  • Understand the issues surrounding running models on mobile/embedded devices and how TensorFlow is being built to address these issues.
  • Understand tradeoffs when designing for various platforms, including GCP and Mobile.


TensorFlow is an open source software library for numerical computation using data flow graphs. Developed by researchers and engineers at Google for the purposes of conducting machine-learning and deep neural networks research, TensorFlow leverages a general computational model that is applicable in a wide variety of other domains, especially for performing large-scale numerical computations.

In this talk we will cover why we built TensorFlow and what were some of the technical challenges in building it.


QCon: Can you explain your talk and perhaps elaborate on TensorFlow?

RAJAT: TensorFlow is an open source software library for numerical computation using data flow graphs. One of the most interesting things happening with machine learning is that once you have a trained model, it’s not about just running it on your server anymore.

More and more we are seeing interest in putting those models on varied devices, including phones and embedded devices. There is a huge set of things you have to consider (many with different kinds of constraints) when you consider these things. For example, there are differences between running on a CPU versus running GPU.

When you go out to these small devices (such as a mobile phone in your pocket, or this little camera or something in your home) the question is how do you design the system, so it works across all of these things?

QCon: If you are designing to go from everywhere from a cell phone all the way to GCP, what are some of the lessons that you’ll be talking about?

RAJAT: When you are designing a system that needs to run across many different platforms, one of the things that has been key (say, if you are going from say a data center to a phone) is how you think about resource management. You really have to think about the phone in advance.

While there are also complexities in the data center, when you go down to something like a phone the issues that you are looking at are maybe power not just compute. I plan to discuss trade offs, such as how to choose different algorithms on different platforms. All of these types of things become very important.

QCon: How deep do you go into these discussions? Do you dive into the trade offs that you specifically made when you were deciding with TensorFlow for mobile and the datacenter?

RAJAT: The discussion has a focus on architectural decisions. How can we plug in different parts in the system, different devices in the back end and still keep the same frontend? How do we layer? This is opposed to talking about the specifics of how do you optimize for a phone, or how do you optimize for a datacenter.

QCon: How you you describe the persona of the target audience of this talk?

RAJAT: This talk is focuses on architects and senior tech leads. Engineers would benefit as well. While it’s not focused on the machine learning side of it for data scientists, they may come away with some interesting information.

QCon: What do you feel is the most disruptive tech in IT right now?

RAJAT: I think machine learning is changing things. Here at Google, we already use it in everything that we do, we’ve seen so many improvements, and I see that really continuing to grow.

Speaker: Rajat Monga

Leads TensorFlow / Principal Engineer @Google

Rajat Monga leads TensorFlow, an open source machine-learning library and the center of Google’s efforts at scaling up deep learning. He is one of the founding members of the Google Brain team and is interested in pushing Machine Learning research forward towards General AI. Prior to Google, as the chief architect and director of engineering at Attributor, Rajat led the labs and operations and built out the engineering team. A veteran developer, Rajat has worked at eBay, Infosys, and a number of startups.

Find Rajat Monga at



Monday Nov 7

Tuesday Nov 8

Wednesday Nov 9

Conference for Professional Software Developers