Page 14 - CCFA Journal - Third Issue

加中金融
CCFF2020/加拿大中国金融论坛 2020
专题讨论 Panel Discussion
主持人：韩玫博士，您拥有机器人和计算机科学的双博士
学位，也曾是谷歌的研究科学家。我好奇什么原因让你从
美国谷歌跳到平安科技？
韩玫博士：大家好，非常荣幸能参与这次的论坛活动。平
安是中国最大的金融和保险公司之一，拥有近 2 百万员工,
而其中大部分是代理。我跳到平安的时候，我爸跟我开玩
笑，说我准备去卖保险。从顶尖大学到研究院，我一直觉
得很幸运。我在谷歌是做面向客户研究的云计算工作。将
人工智能深度学习技术应用到真实的客户需求，这是个很
好的经历。有一次美国卡内基梅隆大学的同学 Jing 找到我，
而他是平安的首席科学家，做的是与云计算类似的汽车损
坏评估项目。Jing 说服了我加入他的组，理由是想用我的
专业知识到车险业务的工作中。他在平安30年的保险工作
经验的基础上，他将所有信息整合起来，而这些重要关键
的信息会使他的系统更高效。平安每天接收三万一千个保
险理赔、每周收到几百万张事故现场照片，这让我很吃惊。
面对这么庞大的数据集，如果我们用人工智能建模的话，
工作将会容易得多。平安的新口号是“金融+科技”，这
给了我一个推动人工智能研究和应用的特殊机会。
主持人：在 2000 年的时候，平安的董事长马明哲曾说过
“互联网浪潮下金融产业不变即死”。他也许是中国金融
行业里最早接受科技的一群人之一。那除了车险事故现场
照片，你每天在研究室还看到了什么数据呢？平安又是怎
么看待传统的保险行业？
韩玫博士：我可以给你举几个例子让你了解我们在平安都
做些什么，特别是在疫情期间。我们通过分析卫星图像对
农作产量进行了预估并预测了此次疫情下中国和北美的损
失。这些分析能帮助我们确定今年哪些农作物的产量会下
降。我们还对远程汽车责任进行评估，这是利用监控视频
进行的一种非接触式评估，系统会通过监控视频自动划分
交通事故主责。这种方法避免了车主争执不休的麻烦，不
用出警，更安全快速地处理车险事故。不仅如此，我们还
研究了通过胸部 X 光片去检测新冠病毒。在给出任何可能
的治疗方案之前，在线聊天机器人会帮助病人识别与新冠
病毒相关的症状。这个系统可以缓解人们的担忧，减少去
医院检查的次数。这些是我们在新冠疫情期间做的一些事。
主持人：谢谢韩玫博士，让我们把金融科技的视角从中国
切换到美国。Touyz 博士，作为第一资本的数据科学家，
对数据研究有如此大的热情，您的博士学位又是研究多元
高斯随机分布，从你的分析来看，在新冠疫情期间，如何
将大数据、数据科学和科技应用到金融行业? 数据科学如
何帮助我们区分和定义用户行为和预测风险？
CCFA JOURNAL OF FINANCE
Page 14
加中金融
Moderator: Dr. Han, you have two PhDs, (one in) Robotics
and (the other in) Computer Science, and you were also a
research scientist in Google. What caused you to transfer
from Google to Ping An?
Dr. Han: Hi everyone, I am very honored to be part of this panel.
Ping An is one of the largest Finance and Insurance companies
in China, it has nearly 2 million employees which most are
agents. When I first joined Ping An, my dad joked that I was
going to sell insurance (rather than research). Going from top
universities to the research lab, I always felt lucky. In my last
year with Google, I was working on their Cloud AI which is
customer facing research. That was a good experience to apply
my AI deep learning technology to real customer requests. At
that time, my classmate from Carnegie Mello University, Jing,
reached out to me. He is the chief scientist of Ping An and was
doing a similar project (to Cloud AI) but for car damage
estimation. Jing convinced me he can integrate my expertise
into the workflow of car insurance business. All these important
and critical decisions made his system more efficient because
he can integrate all this information based on Ping An’s 30-year
insurance experience. It is really eye-opening for me, (for
example) Ping An can accept 31,000 car accident claims each
day. When it comes to vision, they had a few millions of images
each week. It is much easier for us to do real AI modeling given
such a big dataset. Ping An’s new slogan is “finance plus
technology”. This gives me a unique opportunity to push
forward AI research and application.
Moderator: Back in the 2000s, Ping An’s chairman Mingzhe
Ma said that the financial industry either changes or dies
in the new wave of internet technology, and he was
probably one of the earliest to embrace technology in
Chinese financial industry. Other than car images, what
else did you see everyday in your research lab? How does
Ping An look at the traditional insurance industry?
Dr. Han: Yes, I can give you a few more examples of what we
have worked on at Ping An, especially during the pandemic. We
analyze satellite image for yield estimation and predict the
damages for China and North America from the pandemic, the
analysis helps determine which crops will have a worse yield
this year. Another thing we do is remote car liability estimation
which is a no-contact estimation after using just surveillance
videos in China. Based on the videos, the system automatically
determines who may be liable for the accident, so it removes
the hassle of people arguing with each other and eliminates the
need to send the police, which is much safer and faster. We also
do chest X-ray vision detection for COVID-19 symptoms and
host an online chat bot that helps you identify symptoms related
to COVID-19 before suggesting any potential treatments. Our
system gives people peace of mind and lowers the number of
potential trips to the hospital. Those are a few things we did
during the pandemic.
Moderator: Thanks for Dr. Han, now, let us switch from
China to the U.S. from a financial technology perspective.
Dr. Touyz, you are a data scientist at Capital One and you
are so passionate about data research, and your PhD is in
multivariate Gaussian random fields. From your
observation, how can big data, data science and
technology be applied to this industry during the pandemic.
How does data science help us differentiate and define the
user behavior and risk profile?
MARCH 2021