Google's Cloud AutoML uses the company's research and technology to enable enterprises to customize models and tune algorithms with their proprietary data.
Google plans to automate the creation of machine learning models and enable enterprise developers to customize as it aims to use artificial intelligence as a primary use case for Google Cloud Platform. The primary argument will be that Google Cloud Platform's tools for machine learning will produce higher quality models faster.
On Wednesday, Google outlined Cloud AutoML, an alpha effort to give developers technology that can automatically create a machine learning model. Google will start with AutoML Vision and then move through its differentiating categories such as translation, video, and natural language processing.
Fei-Fei Li, chief scientist for Google Cloud AI, said Google has been offering standard AI building blocks, but it became clear that enterprise customers needed to customize models with their own data. "AI and machine learning is still a field with high barriers," she said.
What's the plan to democratize AI? Google's Jia Li, head of research and development of Google Cloud AI, said the automation of technologies such as transfer learning, training, and model optimization will be critical. With automation, Li said customers will be able to improve quality of models much faster. Developers can then turn these models into simple application programming interface (API) calls.
In a blog post, the two Lis noted:
Currently, only a handful of businesses in the world have access to the talent and budgets needed to fully appreciate the advancements of ML and AI. There's a very limited number of people that can create advanced machine learning models. And if you're one of the companies that has access to ML/AI engineers, you still have to manage the time-intensive and complicated process of building your own custom ML model. While Google has offered pre-trained machine learning models via APIs that perform specific tasks, there's still a long road ahead if we want to bring AI to everyone.
Here's a look at what AutoML aims to streamline:
Pricing for this AutoML effort, which starts with AutoML Vision to analyze and optimize images, will be based on API usage and compute. Rajen Sheth, director of product management Google Cloud AI, said customers will pay for the consumption of the API and compute. "We will work with each customer to determine pricing to match their expectations for the problem they are trying to solve," said Sheth.
Google's AutoML Vision will roll out first with early customers such as Disney and Urban Outfitters.
Sheth added that customers will own their data and their proprietary models under the privacy policies of Google Cloud Platform. "This product gives a lot of autonomy back to the customers," said Jia Li.
SELLING QUALITY MODELS
With AutoML, Google is essentially saying quality will be the main selling point for the service. Google's core pitch is that it will use its research and technology to democratize data. Jia Li noted that AutoML is "powered by cutting edge research" and the ability to train models to be production ready faster.
Sheth said these quality models will be created with a simple graphical user interface that shows you the model deployed and the quality. "Quality really matters," Sheth said.
Jia Li said Google has benchmarked its methods compared to existing publicly known techniques to conclude that its models are better and easier to produce.
Even though Google Cloud Platform is simplifying model creation with AutoML it's not all unicorns and rainbows. Companies will need to tag their data and prepare it for AutoML. "In order to create a model for your own purposes, you will still need multiple steps such as model prep, tuning and evaluation as well as iteration. We're providing the technology underneath to remove barriers," Jia Li said.
The launch of AutoML is primarily aimed at developers and independent software providers. Sheth, however, said "I could see this expanding quite a bit to analysts and product people." After all, there are many people in an organization who need to deal with data.
For those looking to deep dive on the models, Google cited the following references on its AutoML post:
Learning Transferable Architectures for Scalable Image Recognition, Barret Zoph, Vijay Vasudevan, Jonathon Shlens, and Quoc V. Le. Arxiv, 2017.
Progressive Neural Architecture Search, Chenxi Liu, Barret Zoph, Jonathon Shlens, Wei Hua, Li-Jia Li, Li Fei-Fei, Alan Yuille, Jonathan Huang, Kevin Murphy, Arxiv, 2017
Large-Scale Evolution of Image Classifiers, Esteban Real, Sherry Moore, Andrew Selle, Saurabh Saxena, Yutaka Leon Suematsu, Quoc Le, Alex Kurakin. International Conference on Machine Learning, 2017.
Neural Architecture Search with Reinforcement Learning, Barret Zoph, Quoc V. Le. International Conference on Learning Representations, 2017.
Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning, Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, and Alex Alemi. AAAI, 2017.
Bayesian Optimization for a Better Dessert, Benjamin Solnik, Daniel Golovin, Greg Kochanski, John Elliot Karro, Subhodeep Moitra, D. Sculley. NIPS, Workshop on Bayesian Optimization, 2017.
Hello, my name is Mohd Azahar. I'm a self-employed Pivrate from the India.
No comments:
Post a Comment