Fireside chat: AI and Big Data - Jason Geng, Modar Alaoui, Xiaojing Dong, Sumit Gupta, Timothy Tang | SVIEF
6:12PM Sep 30, 2018
Well now ladies and gentlemen, we're gonna kick off our panel discussion on the topic of AI and big data. Let's welcome our moderator, Jason Geng, founder and CEO of data application lab.
And our panelists today, Sumit Gupta, VP of big data machine learning. HPC at IBM. Timothy Tang, head of US business Alibaba Cloud. Xiaojing Dong, Head of Marketing science at LinkedIn and Modar Alaoui founder and CEO of Eyeris. The floor is yours Jason
Yeah, thank you. Thank you, everyone. So this is my 30 year to pass the page. This AI and the big data panel I believe with it is the best one we have before. So my name is Jason Geng. I'm CEO of data application lab. We do AI training and consulting for us. I will lead to see it is a fast pan know we had and
I will add to our speakers. To introduce yourself, I will have to start from Modar
Thank you very much. My name is Modar Alaoui, founder and CEO of Eyeris. We do vision AI inside autonomous and highly automated vehicles. We look at the in cabin space, particularly for human behavior understanding along with objects and scene understanding, in order to derive metrics that could enhance safety as well as arguments comfort and convenience.
My name is Sumit Gupta, I lead AI, machine learning and high performance computing in IBM cognitive systems. Our focus is to build a AI development toolkit called power AI, which is based on open source software. So it takes TensorFlow cafe torch and enhances it for ease of use, and auto machine learning capabilities.
Hi, my name is Xiaojing Dong. I'm currently a head of marketing science at LinkedIn. I'm also associate professor of marketing and business analytics at Santa Clara University, where I teach and research in how big data AI technologies can help making better business decisions.
All right. Hey, good morning, everyone. My name is Timothy Tang. I'm responsible for leading the Alibaba cloud business here in the United States. So uh really excited to be here with all of you today.
Wow. You know, we have professors we have, you know, industry leaders today. So, Ai, as, you know, most important technology these days, we saw a lot of, you know, companies like a Binky like IRAs, like IBM like Alibaba, you know, we help, you know, the enterprise to pure there, you know, ai solutions. So, my, okay, the first thing I would like to discuss is, you know, when a company with a corporation, they want to start their AI, you know,
solutions, they want to, you know, design there, you know, ai products. So, what is top of risk of what is chatting, they have to solve? Oh, they they may be face So, I would have to start maybe farm Sumit
Sure, you know, we,
most of my clients are enterprise clients, so other businesses who are in fact starting off on their AI journey, and I have a very simple advice for them, I tell them really, there are three or four steps number one, find a low hanging use case. So low hanging fruit as they say. So always, if you're starting, try to do something simple first, then the hardest part actually, is to get your data strategy together, figure out where your data is, how it's formatted, make sure you label it correctly. And then third is really figured out what insights you want from that data. Right. So that relates to the use case. But the insight will determine how you know what AI technology you're going to use, whether traditional data analytics is good enough, you want to use machine learning, or you want to use deep learning.
Well, go ahead.
Yeah, I second everything summit sad that the only thing is I want to add is certainly something to keep in mind is to add the fact that there has to be social inclusion ethics, in order to avoid any biases, Sumit mentioned annotation or data collection or data collection plus annotation that is a
somewhat in my view, one of the most important things that actually can make an AI or break it. And, and
these are not just like a comprehensive list. But these are some of the things that come come come to mind as of now.
Yeah, I was just going to add that just with a two part keynote speakers, they talked about calculation, plus intuition, and really finding that use case to make a real and I think that's really the power of AI. And when we think about implementing AI in the world, of helping increase productivity, helping increase efficiency, having the data in place, being able to recognize recognize the data and looking for patterns and in be able to solve use case problems. I think that's really the benefit of AI.
Yeah, I just want to add among this industry leaders, so as partially academic, so I sort of added another very big challenges to actually talents. So I work with companies and students. So one thing I realized is that companies need a lot of people who can actually apply to play a lot of these new technologies, but sometimes our education and I'll be competing in life, or even though we learned some, some of these things at school that are completely theoretical, or makes sense, once you go to university or company a year needs a lot of tweaking, and, and learning as well. So, so being utilized these education and industry practice, it's actually a challenge that I see a lot.
Yes. So, um, when we talk about it, you know, Ai, most of us, you know, talk about, you know, Robert, talk about, you know, voice recognition, natural language processing, but it recently technology, cada data science has been applied in many, many, you know, our products, data science is, is a technology on the top of, you know, big data, and we use the machine learning, you know, to
analysis the data is, so I would like to see, how can you talk about it, how the, you know, the data science technology can help with a business to grow to grow their user base to grow their sales, so maybe we can start dr don't,
yeah, so very good classroom. And then it's a question that I got be involved in discussing a lot because I, myself has a PhD in engineering, but I'm a marketing professor in the business school. So I see like it when I was from my personal experience, I was focusing a lot about the technologies. So I got really excited by these new technologies, specially just just now we heard the two fantastic toxin from dark Whoa, whoa, and Dr. Leon, but then when it comes to business, the question is, really, I have this real business problem, I know that your technology can help me, I know that your skill can help me. But But then we see a lot of gaps in there.
So give me give me just waiting cmo because we were told to give some to some use cases it's a wine sample is and I worked with different competencies, everybody care about their sales, right? So I want to increase my sales. But will you talk to a different company, you realize sales means very different things at different stage of the company, or a different units of the company or a different companies. So to be able to end the stand the business, the business advantage, and so the business importance or to define it? What is sales? Exactly? It's actually a tan Indian problem.
So to be able to do that, well, you need to do is you really need to have a good understanding about what is the business so I work with LinkedIn head, a lot of different teams. So I realized that LinkedIn, the well, the biggest thing is, you know, the, the technology or the analytics team, or the data science team, or very close races, Mr. business operations team. So they said that they have a good understanding about what exactly is the talent and then we can tell you that was our data was our technology, what other problems we we can solve.
And the other thing on the data scientists said, the other is really, we have all these really wonderful tools that are built by all these genius people. But then every single tool, every single methodology or models have its limitations, you really need to have a really good understanding of how these limitations because many of these tools, maybe potentially to be able to solve that one question you're asking, which has increased sales, but then given this particular business environments of isn't challenges.
So what are the assumptions which model which methodology, these assumptions are in line or not in line with your requirements, I think that's really it's, it's a challenge. So to do to be able to do that, you really need to have a very good understanding about these tools we, we see a lot of because I work with PhD students as well. So we see a lot of PhD students just grabbing some of these, like these all kinds of tools today are these Google Facebook are building and then just apply it to business questions, it's very easy to make around this area, if you don't understand when these tools are built, the are these assumption they are these limitations. So that's my take out that Thank you. You know,
I can give a few examples from real clients. So
for example, we work with the banks and they have retail banks, right, let's, let's say a bank has 100 retail
outlets. In, in the in in in
California, they have to figure out how much cash to have it every bank. And more often than not, most banks depend on the bank manager ordering cash for the next day. The problem with that is that maybe it's Chinese New Year, right? Maybe it's a Jewish holiday where people give cash out. So often what happens is that the banks are unable to predict the bank manager forgets or doesn't know what the local you know,
festivities might be. or other reasons, it's very simple to build a model using machine learning or deep learning. And that actually can forecast based on historical information, you can just go back to the last two, three years based on your geography, your area, figure out how much cash to have, right? So that's improving operational efficiency. In fact, that improves the Prophet of the bank, the satisfaction of the client base,
another one a, we have a client who was losing, you know, they're very big milk producer. And they will use losing milk crates, you know, the Create in which the make ships, and they didn't know what was happening. Very simple solution, you know,
we worked with a company to put a camera in the docking yard, and they count the number of milk crates getting on a truck in the morning, and
the number of milk crates getting off the truck in the evening. So now, you know, which route is the problem, right? So you can now further debug, okay, why on this route to be lose so many milk crates, so you know, I go back to the point I made earlier, the use cases on savings or improving efficiency should be simple. And you should be looking for simple use cases as you start your AI journey.
So one thing to add is that the way I look at that data science is literally just like what
automation has done to the to the Industrial Revolution, it's literally just automating processes based on a certain type of technology, in this case, the technologies, understanding what the patterns of the data have been over what certain period of time, and then creating models like summit mentioned, that actually could solve these problems, increase efficiency, reduce costs, reduce waste, and all of these things. But it's that the, the, the
prerequisites to have in a great data science is to actually be able to generate creates data. And, and, and the root cause of or, or the roots of that is basically how the deep learning AI has been, has been trained. So you can have a great data science team that are doing fantastic work. But in essence, we cannot discount the quality of data that is generated regardless of how the data science team is doing.
Yeah, well, you know, just add one more point, it's just kind of taking what I just heard, I think the data science is, you know, having the business case, having the, the data and then having a science, I can kind of combine all three, but I think at the end, it's, we still need people to actually look at the data, make a decision and to help us validate continuously. So I think, you know, as we continue to evolve in this, this journey on AI, it's, it's, it will definitely help us make a lot of a lot of decisions a lot faster. But uh, the the bed the idea of continuous evaluation feedback, looking at it to make sure we're making the right decisions based on the right data based on the right artificial intelligence output is super critical for us as well.
Yeah, so there's a nod dot, you know, Ai, data science, you know,
it's changed our life, for instance, you know,
there's a system called a recommendation system, you're almost every day, we got a email of, you know, Mr. Jerome, because our farm snack facts you know, about their products is, is is totally builds on you we prisoner data, right, they build the recommendation for us fee now, even better than my wife, maybe, you know, our boyfriend girlfriend, because they use data, they use AI, they are not assess our, you know, user behavior. But that reason, another question about privacy, right, the now our data, they know what we did at Amazon and Facebook and other social network. So my last question is, so
as, you know, the industry leader in the, you know, research, you know, professors how how we can protect our users state, how we can solve the problem about privacy. So
I'm GDPR has been ready made a lot of impact on in this area as well. But before, even before GDPR, I think this this privacy topic, or even after the GDPR privacy topic, it will be still continuing to be a very hot topic among people who are working in the area of data science and artificial intelligence. So on one hand, you have all these data and you can make, you can help companies and clients or, you know, video consumers to make better decisions and make their life more efficient. But on the other hand, you have too much power to we saw these information about me or other people personally. So
how to balance how to make that balance. So there, there has been a calling for something like GDPR for years in the area of the industry and academic. So I was, I was very happy to see GDPR started I know a lot of people who are working in this area, Neil that a, they had to take on way more complex, complicated jobs senior in everyday work, but I think it does, and this is something that accompanies in Academy has been calling to have some guidance about, you know, from from the third party, or from an ego perspective about how to protect customer or privacy issues before that a lot of company has self control on that area. But, you know, how would this make How would this be more systematic is something that a lot of companies is still struggling with. So I was I was interviewed, that was by some some of these journalists about was Facebook story broke up, so why. So my take on this is, so
that balance. So first thing that balance is very hard to strike. And second thing is the
consumers, or you and I, who are not only working this area, but also as a person, these consumers need to be better educated. So one thing is, they need to wrap up to be aware that your information is public, right, so we heard Dr. Leon talk about the are fantastic, fantastic app. So I was sitting in nearly say, I was thinking, you know, because I make all these lectures is, sometimes I tell the students, it's okay, you can record it. But if you want to record my lecture, you need to let me know ahead of time, right? So I want to be aware about who were my information, what I say, where's that going. So because that's my information, I need to have some control over that. And so I think the number one thing is, consumers need to be better educated to and to be aware of that. On the other hand, I will, I would like to see, you know,
from companies from company's perspective, they need to be also more aware of you may be hurting your consumers or customers or, and yourself at the end of the day, if you don't make the privacy, how you protect your customers privacy, how that information, you need to make that information to be aware to your customers. So that's, thank you.
Yeah, you know,
add to that as well, it's, um, you know, when I think about AI, we need to give our consumers the option to say, I want to be part of this, I want to opt in, just like you said, right? If you want to record my lecture, you need to have my permission. I think one is that, but I think underlying that is really the trust, right? How are you going to use this information. And I think
as long as there's trust in people will know that you can be trusted. And as we as all creators of, you know, ai applications, we have to make sure that underlying our application is the fundament of trust. Without trust, we just won't have people that will believe or use our product. And, and I think trust cannot, is definitely needs to be managed and maintained. Well,
yeah, so, um, I teach many of our, you know, students about, you know, machine learning about you know how to Bute
in tragedy, you know, system, but one of the questions they asked a lot, is about the future of AI. So you see, currently, we see self driving cars. So
we see, you know, Roberts become much, much smarter than it used to be that
we would reach one day, the
AI system itself become smarter than the human beings who designed that, who created that, what do we can do? What's the future of the AI?
I don't think the problem is really who become smarter than the other, it's so much more about what can you do if, you know, if they become smarter? Is it is it like an individual's choice or not, there is no doubt in my mind that AI is already smarter than people, because it can do so many other things as simple calculator does, actually is considered smarter than you are, or any of us here, if we cannot do a simple simple math. And so although that is a, you know, narrow artificial intelligence, but it's what you do with it, or really, and what do you enable it to do and how you have control over it in order to, you know, if it if it exceeds certain limits, what would what do you do about it. And that's where really, you know, a lot of a lot of conversations, about Consortium's being put together that talk about ethics in AI that talks about talk about, you know, standard in how AI can be used and trusts and things like that. And I believe that's really what we should pay more attention to, rather than not focus on or rather than focus on not making it exceeds human intelligence. And, you know,
I think we should look at AI as an assistant to us, right?
How many people are dying this very minute in a car accident around the world. And if we can prevent even a single death by making the cars better at driving themselves, we've done a big service to humanity, right? How many people in the world have access to a cancer specialist? If we can use an AI assistant to enable any doctor to be able to at least give a simple assessment whether someone needs to be sent to a cancer specialist? How can we change the lives of people all over the world? So I think the power of AI is in making humanity better, making us more efficient, giving us all of these things that really are not accessible to everyone.
Yeah, I totally agree with
within the two speakers about how do we leverage the technology of AI, but this also aligned with the topic that we just talked about, about privacy, I think AI potentially will use so the idea of, Hey, I really sees a lot of information, a lot of data to try to make some decisions that will assist us human being, but then how do you use that information, because I've seen some of these really amazing work had a different company has developed using the other data, everybody probably have seen all these movies, talk about, you know, the future net or all this information about how robots can defeat human being.
But if you think about how robot TV human beings really, they leverage all the information that they have, they have information by where you are in a hospital, they have information about where your work, they have information about what credit are us. So I think if we could aware hand developing artists and AI technology to help us living a better and more efficient life, on the other hand, make sure not single company has all those inflammation to leverage all of that to help us
unless they know that they are protecting us. And we don't know where. So that's my fear. Thanks.
Okay. So, um, I think we were talking about, you know,
the AI. So I would like to ask, you know, Sue me to one problem, one question. So, when you guys working on the deeper learning and know, deepening has a very unique technology. So
when the company if they want to use the printer Indian technology, what is here, you know, Chinese for that. So,
deep learning is a method in in the bigger machine learning universe. And it's a method that's particularly effective when you have lots of labeled data. So
three, three important things are lots labeled data, because many times people have a lot of data, but it's unlabeled, or it's hard to label.
A good example, by the way, is medical images. So if you look at medical images, we have lots of images of MRI scans. But the problem is a doctor looked at it says, yep, there's cancer moves on, the doctor doesn't make a circle on the cancer cell and says, this is where I saw the cancer. So you have lots of images, which are cancer, no cancer, but you need much more labeling much more annotation on the data. And deep learning is much more effective. When you have this type of data. It's it's in fact been shown, the accuracy can be 1020, 30% higher than traditional machine learning methods. with deep learning
this, the challenges in this space is collecting the data, labeling the data and then tuning the deep learning models is actually harder than you think. Right. There's a lot of, I would say, science back and black art there, we've been working on making it easier. We have a complete automatic AI software, a complete auto deep learning software for vision. So given a set of images and video, you can actually build a model completely automatically through a GUI by clicking a few buttons. Obviously, it's not as accurate as a team of very good deep learning engineers, but it's actually shown to be much more productive, you know, dramatically improves the productivity of building AI models.
Yeah, I think you know, we have a good talk in the thank you for everyone and thank you for you know, of our audience.
Thank you. Thank you.