- Easy model building with Keras and eager execution.
- Robust model deployment in production on any platform.
- Powerful experimentation for research.
- Simplifying the API by cleaning up deprecated APIs and reducing duplication.
- Low-level control with the Keras Subclassing API, tf.raw_ops, and tf.Module.
- Easy multi-node training on TPU pods and GPUs with tf.distribute.
This year, we are also growing our focus on and commitment to Responsible AI development, to ensure that our models are inclusive, robust, and secure. Two years ago, Google released the AI Principles and unveiled our Responsible AI Practices. Since then, we have been working to build up our understanding of how to build ML systems responsibly and to launch tools that can support developers through this process. Some of these tools include:
- People + AI Research (PAIR) Guidebook
- Fairness Indicators
- What-If Tool
- TensorFlow Federated
- TensorFlow Privacy
- Crowdsource app
- Model cards
- TensorFlow Constrained Optimization (TFCO)
WHAT SHOULD I HACK?
We want this challenge to be both about creating something great with TensorFlow and creating something that has the AI Principles at heart.
What do we mean by that?
Use our official release of TensorFlow 2.2 to do something nifty: build a model, a mobile or web application, an art installation, or something else entirely! As you build, at different moments of your product development process, ask questions related to fairness, privacy, and security. For example, when collecting data: What are the privacy considerations? Does your data represent the diversity of your users? You can also use our interpretability tools to better understand and debug your model’s performance. To learn more about our Responsible AI practices, check out this link, and read through the tools above.
We are still learning and building up both our understanding and toolkits around Responsible AI, so please reach out to the Github with questions and insights! We’ll learn together.
Grand Prize (5)
Five participants will receive (1) the opportunity to virtually present their solutions to the TensorFlow engineering team; (2) a limited-edition TensorFlow swag kit, and (3) have their solutions featured on TensorFlow social media channels.
Submitting to this hackathon could earn you:
Developers of all ages, backgrounds, and skill levels are encouraged to submit projects. Teams may have between 1 and 6 participants. Participants are encouraged to expand the scope of an existing TensorFlow 1.x project to TensorFlow 2.2, to migrate a historic TensorFlow 1.x project to TensorFlow 2.2, and continue work on it; or to create an entirely new software solution using TensorFlow 2.2.
Applications may submit projects that integrate with other third-party SDKs, APIs, and services, provided the participant is authorized to use them. There are no restrictions on cloud service providers, operating systems, or hardware platforms. All participants will be obliged to work under the TensorFlow Code of Conduct.
Note: government officials, corporations, and employees of Google or DevPost are not eligible to win prizes, but may submit projects. See Responsible AI with TensorFlow 2.2 Challenge terms for further restrictions.
Build a functioning Tensorflow 2.2 based solution, and tell us about how you leveraged the Responsible AI practices as you did so.
- (Optional) Submit a 2-5 minute demo video hosted on YouTube, Vimeo, or Youku. Your video should include a demo of your working application, and any Responsible AI considerations and approaches.
- Please submit at least one image or screenshot of your solution.
- Please submit a PDF document discussing the Responsible AI concerns, any tooling you used, any approaches you took to address these concerns, and any challenges you faced in the process. If you have any requests from Tensorflow tools, let us know that as well!
- Make sure all of your code has been uploaded to a public repo on GitHub or another public repository, and that a link to the repo has been included in your application.
All projects must be submitted by May 11th, 2020, at 11:45PM Pacific Time. Judging will take place during Google I/O, from May 11th through May 15th. Contest winners will be informed of their project's status on the evening of May 19th.
Product Manager (ML Fairness)
Product Manager (TensorFlow)
Kemal El Moujahid
Director of Product Management (TensorFlow)
Engineering Director (TensorFlow)
Product Marketing Manager (Google Research)
Engineering Director (Google Research)
How original and innovative is the product? Is there already an existing model or software application with similar functionality? Does this project integrate deep learning in an innovative, unexpected way?
Does the product integrate a large number of components? Is it widely useful, and scalable? Is the software application or model usable in a production setting? Did it incorporate responsible AI best practices?
What kind of social or business value could this product deliver? How does this software application use deep learning to make the world a little bit better, and to positively impact people’s lives?