Yael
Karov

AI For Assisting in Task Completion
Director of Engineering at Google
Yael Karov Zangvil

Yael
Karov

AI For Assisting in Task Completion

Director of Engineering at Google
Yael Karov Zangvil

Bio

Yael Karov is a Director of Engineering at Google, leading next-generation AI products for mobile phones. Before joining Google, Yael managed Cortana language understanding and dialog management for Microsoft. Before that, she founded and managed Ginger Software, the world leading grammar checker, as well as its business unit for personal assistant language understanding that was acquired by Intel. Yael was the CTO and VP R&D of Rosetta Genomics – her team discovered the majority of microRNA genes with ML technology and released 5 FDA-approved cancer diagnostic products based on the discovered genes, leading to the company’s IPO. Yael is the inventor of 53 patents. She has Ms.c in computer science from Weizmann Institute of Science.

Bio

Yael Karov is a Director of Engineering at Google, leading next-generation AI products for mobile phones. Before joining Google, Yael managed Cortana language understanding and dialog management for Microsoft. Before that, she founded and managed Ginger Software, the world leading grammar checker, as well as its business unit for personal assistant language understanding that was acquired by Intel. Yael was the CTO and VP R&D of Rosetta Genomics – her team discovered the majority of microRNA genes with ML technology and released 5 FDA-approved cancer diagnostic products based on the discovered genes, leading to the company’s IPO. Yael is the inventor of 53 patents. She has Ms.c in computer science from Weizmann Institute of Science.

Abstract

In the recent decade there is a massive work on how AI can assist the user with task completion over mobile devices, starting with personal assistants like Google Assistant and Cortana, then the revolution of chat bots for reservations, customer service, and topic exploration, and recently operating system AI features that can assist the user with their task completion.

 

In this talk I will describe the science components for a task completion assistant, including: understanding user intent over natural language queries, knowledge graph entity classification, sentiment analysis, dialog management, conversational data collection and annotation, and evaluation.

 

I will also describe how Federated Learning enables mobile devices to collaboratively learn a shared model for task completion while keeping all the training data on device. Hence, it enables learning from user mobile behaviour and feedback loop instead of learning from annotated crowd source data.

Abstract

In the recent decade there is a massive work on how AI can assist the user with task completion over mobile devices, starting with personal assistants like Google Assistant and Cortana, then the revolution of chat bots for reservations, customer service, and topic exploration, and recently operating system AI features that can assist the user with their task completion.

 

In this talk I will describe the science components for a task completion assistant, including: understanding user intent over natural language queries, knowledge graph entity classification, sentiment analysis, dialog management, conversational data collection and annotation, and evaluation.

 

I will also describe how Federated Learning enables mobile devices to collaboratively learn a shared model for task completion while keeping all the training data on device. Hence, it enables learning from user mobile behaviour and feedback loop instead of learning from annotated crowd source data.