Adva
Zair

Practical Deep Learning: From Paper To Feature

Lightricks

Adva
Zair

Practical Deep Learning: From Paper To Feature

???

Bio

For the past 3.5 years, Adva has been working at Lightricks as a research ML engineer. Adva has a BSc in Computer Science & Computational Biology and an MSc in Computer Science from the Hebrew University (cum laude). During her studies, she worked at a part time job developing an ancient script classification model for the Computerized Archaeology lab in the University.

Bio

For the past 3.5 years, Adva has been working at Lightricks as a research ML engineer. Adva has a BSc in Computer Science & Computational Biology and an MSc in Computer Science from the Hebrew University (cum laude). During her studies, she worked at a part time job developing an ancient script classification model for the Computerized Archaeology lab in the University.

 

Abstract

In this talk, I will describe the life cycle of a deep learning (DL) research project – starting with a vision of a designer, and all the way to creating a feature used by millions of users. We will discuss the challenges of building DL models under mobile device constraints, how we tune the model to our needs, and I will provide some practical tips for training the model. These topics will be demonstrated using some features I developed, such as style transfer, saliency segmentation, and clothes segmentation.

 

Abstract

In this talk, I will describe the life cycle of a deep learning (DL) research project – starting with a vision of a designer, and all the way to creating a feature used by millions of users. We will discuss the challenges of building DL models under mobile device constraints, how we tune the model to our needs, and I will provide some practical tips for training the model. These topics will be demonstrated using some features I developed, such as style transfer, saliency segmentation, and clothes segmentation.