APIs Beginners Data Science Deep Learning Machine-LearningSee in schedule
The talk will showcase the main features of this new TF version that should appease the needs of the Researcher and Practitioner alike. A focus will be put on:
- Keras as the new standard API
- Eager Mode & Graph
- Production Pipeline
- Extended Ecosystem
Prerequisites: Basic Python, Exposure to a deep learning framework is advised but not necessary
If you are even remotely interested in machine learning, deep learning or data science, you have probably heard of TensorFlow. Released publicly on November 9, 2015, by the lovely people of Google Brain, TensorFlow is not only one of the most active (and starred) Open Source projects on GitHub but most importantly the most popular deep learning framework.
Since its inception, TensorFlow has always had a simple set of goals in mind: performance, scalability and a relatively straightforward path to production (at Google scale).
While over the year TF achieved each one of these goals becoming the defacto gold standard of deep learning framework, ease of use was not one of its selling points, but actually, it's Achille's heel. Until the coming of Keras, the TensorFlow programming experience was not a pleasant one and even then the often too high level, and magical Keras was not enough for more complex tasks.
However, what TF lacked in that area it more than made up with raw performance and most importantly an ecosystem of tools, libraries, and services that made deploying models in productions a straightforward experience.
Things have however changed in these four years; the deep learning world has seen a Cambrian explosion in terms of growth, new APIs, competing frameworks and an exponentially larger audience demanding from TensorFlow a more modern and Pythonic API much like the one offered by PyTorch its main competitor.
The core team has listened to its users and has given us TensorFlow 2.0.
TensorFlow 2.0 has done away with an old, crusty, poorly designed and cumbersome API by adopting Keras as the default model specification API while also learning a thing or two about eager execution from Pytorch. The result is an almost an entirely new framework when it comes down to usability while the production side of things has only got better and better over time thanks to an ever-expanding universe of supporting projects.
Type: Talk (45 mins); Python level: Beginner; Domain level: Beginner
Lover of 🐧🐧. Pythonista 🐍. Machine Learning Engineer 🤖. Mad Scientist. Evil Mastermind. Walking Beard. Tinkerer. Nerd. Tech junkie.
Programming turned the tide of a crippling, panic-attacks inducing depression caused by a profound unsatisfaction with my choice of an academic career. During 2016 and 2017 I developed a burning passion for Python, robotics, Linux, Open Source/Hardware/Data/Science, and all things Machine Learning, I devoted myself day and night to learn the craft. My passion never waned and my drive toward knowing more grows each day.
Currently employed as a Machine Learning Engineer at https://zuru.tech, where I lead the research effort on GANs (and everything else relating to either the generative or the adversarial world of Deep Learning), help with Computer Vision and act as the Supreme Overlord of the Data Pipeline that feeds our AIs.
Consultancy --> https://ubik.tech
Personal Site (coming soon): https://ubik.dev