Supercharge Your Career with TensorFlow

Keeping up with advances in machine learning can be overwhelming. The landscape changes daily, and it’s not obvious which new trends are just fads and which are radical shifts in terms of what’s possible. It can feel like you need a math PhD to really understand what’s going on.

Have you heard of TensorFlow but don’t know where to start? Maybe you know that TensorFlow has gathered an immense community of engineers, empowering them to solve data-driven machine learning problems — yet it’s not clear to you exactly what TensorFlow is good at or what it’s limitations are. How can TensorFlow move your work forward?

That’s where this workshop comes in. I’ll help you feel confident:

• Identifying which machine learning problems are a good fit for TensorFlow or neural networks.
• Teaching your peers how to architect, code, and train the right shape of neural network for their problem — from scratch.
• Knowing how to dramatically speed up TensorFlow’s training time for any system by allocating and taking advantage of GPU-based computation.
• Starting with simple networks using a straightforward API and having the ability, if necessary, to peek under the hood and achieve fine-grained control of your solution using lower-level TensorFlow functionality.
• Grasping the intuitive, motivating concepts behind large-scale machine learning decisions — in other words, how to fit a specific approach to your specific problem — and why some solution setups work well for certain problems.
• Understanding major neural network architectures, such as CNNs (convolutional neural networks), RNNs (recurrent neural networks), and MLPs (multi-layer perceptrons).
• Working with domain-specific best practices as you and your team integrate machine learning sophistication into your process.

TensorFlow is more than a library: it’s a city-sized ecosystem of interfaces and tools that’s evolving quickly and in fundamental ways. There’s plenty of info out there about TensorFlow, but it can be hard to know which concepts are truly useful. TensorFlow 101 will teach you what you really need to know — the fundamentals that practitioners use to solve real-world problems.

You might have seen Medium posts or YouTube videos explaining key Tensorflow concepts. But I’ll bet you walked away feeling a bit unsure about how all the pieces fit together. What is the unifying theme of TensorFlow, beyond something abstract like neural networks? What is its design philosophy? What does it do well? Where does it draw the line between what you control and what it abstracts away? What is the conceptual framework behind a well-designed TensorFlow application?

If you’re a hands-on learner, perhaps you’ve tried downloading a pre-existing TensorFlow script to see if using it, or making slight adjustments to it, can help you learn or solve a particular problem. That’s a great way to get started, though TensorFlow has its share of idiosyncratic and confusing error messages. These are exactly the type of issues that I’ll bring up and clarify in a way that cookie-cutter, sanitized code can’t. The lab portion of this workshop is the perfect place to get stuck — I’ll be right there to give you the 5-second insight that even a good ol’ Google-of-the-error-message can miss.

I designed this workshop for software engineers who don’t already have a machine learning background. The only background you need is a working knowledge of Python. I’ll cover the key theoretical ideas, at a high level, as we go through the material.

TensorFlow is a tool for machine learning, just as the microscope is a tool for biology. You need to know something about biology to really appreciate the power of a microscope; similarly for machine learning and TensorFlow. Because of that, I’ll provide an intuitive introduction to the main concepts of deep learning , which is the use of neural networks within machine learning. It’s like learning a bit of biology when you’re being taught how to use a microscope for the first time.

If you’re worried about not having enough background in math, fear not: this workshop is designed for all Python engineers. You don’t need to know what gradient descent is or understand matrix calculus. This is a programming workshop and not a math workshop. I’ll teach the few key mathematical ideas you need to know to really grasp what TensorFlow is all about.

My teaching philosophy is to present an intuitive and concrete grasp of the main ideas. I’ll visualize ideas, speak about technical topics in plain English, and provide practical, hands-on experience that maps directly to real data and the kinds of challenges you’re likely to see in your own work. I’ll guide you through the TensorFlow landscape so that you know how the various API hills and valleys fit together, and you won’t feel lost staring at a particular leaf, unaware of the forest. Perhaps most importantly, I’ll tell you why things work the way they do in TensorFlow and explain how you can adjust your approach for your specific data and the problems you're grappling with in your work.

You’ll walk out of this workshop feeling like you know what TensorFlow does, why it works the way it does, and the kinds of problems it can help you solve. I’ll empower you to start with data, ask the right questions, and explore TensorFlow-based answers with fluency.

Key Topics Covered

In this daylong workshop, we’ll:

• Build an intuitive understanding of what neural networks are, as well as a formal way to express them as functions.
• Learn how TensorFlow sees neural networks and why it’s designed to see things that way.
• Examine the key aspects of TensorFlow’s implementation of the keras API — a simple yet surprisingly powerful API for quickly building, training, and testing neural network models.
• Explore the lower-level, fine-grained core TensorFlow API for defining arbitrary computation models.
• Talk about the main ideas behind CNNs and RNNs, and the kinds of problems where each is most applicable.
• Discuss some of the common theoretical difficulties that arise in building and training models — such as overfitting — and various methods to diagnose and tackle these challenges.

Instructor

Tyler Neylon has over a decade of experience developing machine learning systems for real-world applications. He built Google’s initial sentiment analysis engine in 2006. Tyler also created robotics simulations for Open AI, developed natural language processing techniques for Primer AI, and co-founded the data science infrastructure company Zillabyte through Y Combinator. He wrote the O’Reilly book Creating Solid APIs with Lua and has three patents focused on sentiment analysis. Tyler holds a Ph.D. in Applied Math, specializing in machine learning algorithms, from NYU.