Introduction to Hilbert AutoML with TensorFlow Extended (TFX) at Yahoo! JAPAN

SHIN-ICHIRO OKAMOTO (Actapio f.k.a. YJ America)

4:10pm–4:50pm Wednesday, October 30, 2019

Location: Grand Ballroom C/D

Production pipelines

Download slides (PDF)

Who is this presentation for?

Data scientists, ML engineers, researchers, and ML ops and DevOps practitioners

Level

Intermediate

Description

Yahoo! JAPAN provides more than 100 services, which include advertising, search, ecommerce, auction, finance, maps, news, travel, video streaming, gaming, cloud service, data business, etc. The company’s data scientists and engineers have different skill sets, and it was necessary to develop an AI framework to meet their individual needs so everyone can make deep learning models more easily.

In order to realize production-level AutoML, the company introduced a complete TFX YAML configuration interface. As a result, it’s now possible to optimize hyperparameters, input feature combinations, input cross-feature combinations, embedding sizes, model types, model architectures, model size, and more at the same time. It would be difficult to achieve all of these optimizations without a TFX configuration interface.

Hilbert is an AI framework that works with TFX to provide AutoML to create production-level deep learning models automatically. Hilbert is currently used by over 20 services at Yahoo! JAPAN.

Shin-Ichiro Okamoto explains how to achieve production-level AutoML and explores service use cases at Yahoo! JAPAN. He also details the company’s goal to make Hilbert open source.

Prerequisite knowledge

A basic understanding of machine learning and deep learning model training workflows and optimization techniques (useful but not required)
A working knowledge of TensorFlow (useful but not required)

What you'll learn

Understand how to create deep learning models applied to real services
Discover the use cases of deep learning and machine learning at Yahoo! JAPAN

SHIN-ICHIRO OKAMOTO

Actapio f.k.a. YJ America

Shin-ichiro Okamoto is the vice president of data science division at Actapio. Shin-ichiro works at Actapio on behalf of Yahoo! JAPAN. He develops AutoML with TensorFlow Extended and leads the AI research and development of Yahoo! JAPAN. Previously, he was the divisional chief technology officer in the Data and Science Solution Management Group at Yahoo! JAPAN.