AI公开课：19.05.22 Aya Soffer—IBM AI Tech VP《Advanced, scalable, and trusted AI 》课堂笔记以及个人感悟

本文是IBM AI公开课课堂笔记与感悟。课上IBM AI技术副总裁Aya Soffer与雷明教授就NLP、计算机视觉、中美深度学习差异等问题展开问答。还介绍了深度学习革命、AI的演变、企业AI议程等内容，提及AI信任、模型生产等挑战。

Introduction
Aya Soffer is Director of Information Management for Analytics Research in IBM. In this role, Aya is driving the world-wide software research strategy in information management for analytics, ensuring IBM's position as a premier software research organization in this area and delivering Research technologies to IBM's software products and customers. In the Haifa Research Lab, Aya is the Department Group Manager of the Information and Interaction Technologies department that focuses on inventing and developing cutting edge technology that connects people and systems to information. Many of these innovations are already included in IBM product and services offerings. IBM’s product groups and customers to drive Research innovation in to the Market.

The Session of Question and Answer

Live Pictures

The PPT of speaker

The Session of Question and Answer

I am trying, please wait……

Professor LeiMing：from your own view，please talk about the development of NLP in the future
Aya Soffer VP：word embedding ,but deep learning is not enough.
Professor LeiMing：about the bert of google，so, computer power will be play an import part in the future?
Aya Soffer VP：many many people ask, In fact ,it is diffficult,
Professor LeiMing：ok,let‘s, talk about computer vision，
Aya Soffer VP：it has many many application，CV is popular, Tuling test. simple case from AI，but we can do many amazing thing ,such as ALibaba. So，I think our work is just starting.
Professor LeiMing：AI, How far and How long. please talk about the difference of DL between the US and China. In China, we have a large number of users and a large number of data sets.
Aya Soffer VP：In fact,I don‘t know, may be 50years？ haha. We have different cultures.
Professor LeiMing：this question is from student. Talk about Alphazero
Aya Soffer VP：Alphazero，More data needs to be fed, and its performance will get better and better. It's amazing.
Professor LeiMing：another question, Lawyer's position will be defeated? And the emotions of AI, How will it develop?
Aya Soffer VP：AI can provide better service for customers. The emotions of AI , some way to go.

Personal views：NLP will shine brilliantly in AI field.
Note: the text of the above dialogue is edited for the blogger to summarize. If slightly different from the original, please refer to the original recording. The time is pressing, if has the mistake, welcome the netizen comment to point out, the discussion.

Live Pictures

I am trying, please wait……

The PPT of speaker

I am trying, please wait……

AI is …
The deep learning revolution

Deep Neural Network
Large Data for Training
GPU Hardware Acceleration

The evolution of AI

Narrow AI
Broad AI
General AI

AI Agenda for the Enterprise

From narrow to broad AI
Operationalizing AI at scale

Who won? Two measures
Stance change

offline Analysis VS online Analysis

Calim Detection

、

Three major capabilities that power Project Debater

Data-driven speech writing and delivery
Listening comprehension
Modeling human dilemas

Potential Applications

Many industry problems require AI for vision and video
Challenge in video comprehension for the enterprise

Few-shorts and One-shot Learning

kids learn from a single example
Get machines to learn more like humans

Moments in Time Dataset
A large-scale dataset for recognizing and understanding action in videos
Activity Recognition

How can you put your AI models into production at scale?
Existing approaches leave two important questions unanswered

AI System:Automation tool(Power AI Inference Engine)for deep learning accelertor generation and optimization
AI AutoLearning-High Efficient AutoML

Trusting AI
Go beyond accuracy and performance metrics, to instrument AI models and systems with dimensions of trust, making
them fair, explainable, safe, accountable, and compliant with human norms and values.
What does it take to trust a decision made by a machine