Top 5 from the TensorFlow Dev Summit 2020

Key Takeaways

greetings tensorflow enthusiasts I'm Paige Bailey for the developer show and I'm at the tensorflow dev summit where members of the tensorflow engineering team just shared major release announcements with the world this past year has been a busy one for the tensorflow team we've launched courses on Udacity and deep learning AI we released many new products like tensor flow graphics MLA R and TF Enterprise and we got to meet so many of you at events like the tensorflow roadshows tensorflow docs prints google Summer of Code and google code-in this year is shaping up to be just as exciting and now without further ado the tensorflow dev summit top 5 first up the release of tensorflow 2.2 we've launched gif 2.0 last year to very positive feedback from the community as a recap we've also added TPU and multi worker GPU support is part of TF 2.1 you can now scale from single node multi-gpu to multi node multi-gpu to TPU pods with just a single line code change we've also been excited to see the amazing libraries that you've built with tensorflow 2x and the speed at which developers are migrating their existing tools but we heard your feedback loud and clear tensorflow 2 is fast but you wanted it to be even faster and to have tools to better understand your performance that's why tensorflow 2.2 was created with an emphasis on performance compatibility and stability we're building on the momentum from last year and have many improvements under way including a rewrite of tensor flows internals and de C++ to make it even faster we're also creating a numpy like eager core and including all the low-level operations you need as part of TF raw ops we're also excited to release a performance analysis framework that lets you inspect and intuitively visualize the execution timeline of accelerators like GPUs GPUs and it has low overhead so on GPUs typically about 2% overhead when profiling or tracing and it's highly customizable profile when you want select duration and desired performance counters simply add a build dependency and you're good to go be sure to share your visualizations on social media with the hashtag powered by TF coming in at number three for expanding the way students and researchers can get involved in the tensorflow community everything from google Summer of Code applications open on March 16th and credits for TP use as part of TF research cloud there's so many ways for tensorflow to help you with your academic work we know that machine learning is a powerful way to solve challenging real-world problems however machine learning and AI raised many questions around the best approach to building a system that is both an inclusive and secure that's why number 4 on our list is tensorflow suite of responsible AI tools for example fairness indicators can help you evaluate simple but critical fairness metrics across thresholds of models the what-if tool directly integrated into tensor board can help with interpretability providing a rationale for why a prediction was made using textual and visual components of the data and tensorflow federated now updated for tensorflow 2 is a library that helps with training data on device we're excited to work on this important part of the tensorflow ecosystem with all of you the tensorflow community number 5 on our list you can share your work with the broader tensorflow ecosystem and get recognized for your contributions first up we're announcing the responsible AI dev post challenge which gives you the ability to use the machine learning fairness tools I just mentioned to create equitable ái'm office we're delighted to announce that Kaggle competitions now support tensorflow 2.2 and teepee use be sure to check out the new kernels and try migrating code from CPU to GPU to TPU for all the announcements and updates from the tensorflow dev summit make sure to check out the links in the description below I'm Paige bailey for the developer show thanks for watching [Music]