Last week’s TensorFlow Dev Summit 2020, hosted as a live streamed virtual event due to safety concerns around COVID-19, offered an overview of the meanwhile extensive ecosystem of TensorFlow.
In a blog post, Developer Advocate Laurence Moroney recapped what was presented at the summit, so let’s take a closer look.
SEE ALSO: TensorFlow.js now has a WebAssembly backend
TensorFlow Hub
For example, work on TensorFlow Hub is continuing. TensorFlow Hub was launched in March 2018 as a repository that hosts pre-trained machine learning models developed by Google and the AI company DeepMind, which was acquired by Google in 2014.
The repository allows you to search among more than 1,000 models using different criteria such as domain (image, text or video) or model format (TF.js, TFLite or Coral).
An introductory blog post offers further information on TensorFlow Hub.
Protecting AI Solutions From Attacks
Alexander Polyakov (Independent Researcher)
AI for Decision Makers
Uri Eliabayev (Machine & Deep Learning Israel)
Reinforcement Learning and Imitation Learning Workshop
Dr. Neda Navidi (AIR (AI Redefined))
TFX and Google Cloud AI Platform Pipelines
At TensorFlow Dev Summit, the new Google Cloud AI Platform Pipelines were introduced, which should make the process of bringing TensorFlow models to production easier.
In order to use the new Google Cloud AI Platform Pipelines, users must install Kubeflow Pipelines with TensorFlow Extended on a Kubernetes cluster. TensorFlow Extended (TFX) is an “end-to-end platform for deploying production ML pipelines,” which was launched in 2019.
Here’s what a production system built using TFX looks like:
SEE ALSO: Kubeflow 1.0 brings cloud-native machine learning to Kubernetes
TensorFlow Lite
Another area of TensorFlow, the open source deep learning framework TensorFlow Lite for on-device inference, is receiving some updates as well. Canary Channel participants will soon be able to use TF Lite with Android Studio, allowing automatic generation of Java classes for TF Lite models by dragging and dropping models into Android Studio.
Model authors can now also provide metadata, and the new Model Maker allows the fine-tuning of pre-existing models. Other updates were made regarding performance on mobile phones and usage on Apple devices.
TensorFlow Developer Certificate
And, last but not least, you can now prove to yourself—or rather to prospective employers—how advanced your TensorFlow skills are by obtaining the TensorFlow Developer Certificate. This doesn’t come free of cost as participants must pay $100 to take part in the exam.
The program does, however, offer a limited number of stipends to “widen access to people of diverse backgrounds, experiences, and perspective.” The stipends apply to educational material as well as exam costs.
See the website for further details.
The post A look at the TensorFlow ecosystem in 2020 appeared first on JAXenter.
Source : JAXenter
