Data Science and AI for COVID-19 with Professor James Zou and Dr. Irena Fisher-Hwang
36:35 Share

Hello and welcome to Skyner Today's Let's Talk AI podcast where you can hear from AI researchers about what's actually going on with AI and what is just clickbait headlines. I am Andrey Krenkov, a third year PhD student at the Stanford Vision and Learning Lab and the host of this episode.

On this special interview episode, you'll get to hear from Stanford assistant professor James Zhou, who works on a wide range of problems in machine learning and especially applications in genomics and computational health. He has also taught classes such as value of data and AI, algorithms of advanced machine learning, deep learning in genomics and biomedicine, and the brand new data science and AI for COVID-19, which is just being taught this quarter at Stanford.

He also enjoys writing and journalism and has been a common alumnus for The Budapest Sun and The Chronicle. We will also hear from Dr. Irina Fischer-Huang, who has a doctorate in electrical engineering at Stanford with research centered on bioinformatics, image compression, and science communication, and is currently helping to teach with data science and AI for COVID-19 class.

She is also currently a graduate student in the Masters of Journalism program at Stanford University, focused on tech culture, explainers about tech and diversity in tech. Thank you so much, Professor Zhou and Dr. Fishing Huang for making the time to be on this episode. Thank you, Andrea. Thank you very much for having us. Thanks for having us. Great. So let's go ahead and dive straight in by talking about this data science and AI for COVID-19 class.

The course description says it is a project class that investigates and models COVID-19 using tools from data science and machine learning. At the core of this class will be projects aimed to create tools that can assist in the ongoing global health crisis. So what was and is your motivation for creating and teaching this class?

Right. So my research group actually works on, as you said, works at the intersection of machine learning and biotechnology and healthcare. So we've actually been working on several COVID-related research projects for the past few months.

And we're just starting to actually get more and more projects, more and more ideas for different projects, both from within our group and also from our collaborators at Stanford Medical School, at the hospital, from pharma companies, from biotechs. And we thought, OK, so we can't really do all of those projects ourselves, but we have this amazing resource here at Stanford, which is really we have just amazingly talented graduate students who are very generous and very passionate.

And everyone's really trying to help. So we thought, okay, so maybe the best way to really get this going is to try to essentially crowdsource these projects and to get a lot of students involved. So there are basically two main motivations for organizing this class. The first one is to get the students to work on these research projects, which we think will have a high impact in real time. The second one is that we also want to, through this class, create a set of

educational materials around COVID-19, especially aimed towards data scientists and AI researchers, because we think that those materials will be really useful for the broader public. And with the help of ARENA, we're also putting these materials on YouTube, online. Oh, that's amazing. And yeah, ARENA, maybe what's your take on this class? Yeah, it's been a really interesting experience. And I think it is coming at a really useful time because

I think right when James contacted me, it was the first few weeks of shelter in place orders in the U.S. And it was there's just this overwhelming sense of like, we don't know what's coming next. And there's also, I think, a lot of confusion about what information was coming out.

So for me, it's been really helpful actually to even just sit in on the lectures as a TA and to sort of get the very high level overview about the problems hospitals are facing when it comes to sourcing PPE or how epidemiological models work, stuff like that. So I've personally enjoyed it very much. And I think it's also wonderful to see academia applying its skills towards such a current and pressing problem.

Yeah. And launching off of that point of applying its skills, James, you said that part of the motivation was to actually get more grad students to work on projects. So it sounds like there's a lot to do. So what kind of projects and work and problems is the class kind of centered on?

Right. So the class is actually very interdisciplinary. So the students in the class are mostly graduate students. They're mostly PhD students. So we have some Stanford undergrads and master's students and MBAs. And they're coming from really diverse backgrounds. So everything from economics to communications to their MD, PhDs.

And this is great because I think there are so many complex components to COVID-19, to the crisis. So we want to have some projects that are really looking at the fundamental biology of the virus. So basically, what is the virus actually doing to the human cells? What is the genetics of the virus? We also want to have some projects that are really looking at the broader social impact of the pandemic. So there are some students that are actually looking at Twitter data to see how it's

misinformation spreading on Twitter, how does that actually change the compliance to shelter in place? We have some students looking at mobility data that you can gather from cell phones to see how is the mobility actually changing in every county in the US and how does that actually change the spread of the virus, the spread of the cases in different counties.

We're also having various projects that are working with our collaborators in, for example, pharma companies like Genentech, Roche, where the students are actually really working with their teams on prioritizing where to send potentially useful medical treatments. So it's quite a diverse set of projects and the students are really applying their own unique background and interests to make very substantial contributions in the ongoing global efforts.

Wow. So the hope is actually that not only is it educational, but these projects will ideally, at least in some cases, have an impact beyond the class on the actual efforts out there, right? Absolutely. Yeah. So then this is what we sort of stress to the students in the first lecture, right? Because in some sense, everybody is basically volunteering their time. So we're very grateful for the students and for the TAs like Arena to contribute and

And moreover, we would like the output of the project, output of the class to really go beyond standard course projects. So we could have right papers, it could be blog posts. And in parallel, we also love for the students to really push these projects out into real tools, which would be websites, dashboards. And those to work with other companies are actually really trying to implementing and deploying the tools that students are building in real time.

And that's really cool. I think that's something that is really cool about Stanford in general, that a lot of classes have such project centeredness and trying to do really substantial things. Irina, I'm curious, how has being a TA for this class and trying to help other projects been going so far?

Yeah, it's certainly been interesting. Um, I think prior to this quarter, I had used zoom only handful of times. Um, and then all of a sudden I was helping to, uh, administrate, I think like up to 90 person zoom call. And, and so it's, it's, it's been a steep learning curve. Um,

But it's also been really interesting to see how quickly we can pivot to the sort of online learning model. James has done a fantastic job of lining up really interesting speakers who are able to give really compelling presentations with the help of their slideshows. And I think that in that sense, it's been going pretty well. I think that...

The main challenge, though, is because it is a project oriented class. It's a little hard, I think, to do a project when you can't physically be together. I mean, projects or group projects are challenging enough when you have a number of very bright students, all with different talents.

But when it comes to sort of finding a good time when some of them may have gone home and are in different time zones and also lacking that sort of face-to-face interaction, it's just a little bit more difficult. So I think that it's been important to try to keep a pulse on groups a little more and just to be aware that some students may be facing challenges that others aren't. That makes total sense. As a researcher trying to stay productive myself with some collaborators, I

I have had to learn some new tricks to keep myself motivated. And to that last point you had about online education, I wonder, has putting together this class and teaching it influenced your overall outlook on online education and how much we can rely on it in addition to more traditional in-person classes?

I can't say that I had particularly strong opinions about, say, massive online courses. I personally, so I am also taking classes right now at the same time. And I will say that I do miss very much the sort of

like casual face-to-face interactions you can have when you're in person. It's, you know, it's hard to not be able to bump into your classmates in the hallway and have impromptu discussions about things that you don't discuss in class. And of course, I think the main challenge in something like video conferencing for a classroom setting is that you can't really have conversations in parallel. You can't really do learning in parallel, right? Like sometimes in person, you can have a main discussion going on and a few satellite discussions, right?

So I think it is challenging to deal with those issues. But I am also optimistic. Like, I think that I think that the students have risen to the occasion and and then professors like James have done a good job in setting up a compelling format. Yeah. And yeah, Professor Joe, how about you? What have your feelings been on the online education?

Yeah, that's a great question. This is actually my first class that I'm teaching online. So all of my other classes, the ones that you've mentioned, they've all been basically taught in person. And I really love the interactive nature, as Irina said, the personal interactions that we get from the in-person classes. The other part that makes this particular class very different is that it's really interactive.

pulled together at the very last minute. So it's actually a pretty funny story that basically, I think on March 16th, right, like 10 p.m. at night. This is why I first thought, OK, so maybe it's not so crazy to try to do something like this together, put this class together. And the quarter starts like late March sometime, right? That's right. Yeah.

So that very nice. I wrote an email to the CS department to Mehran, who is in charge of the course, the course material. And he very kindly wrote back the next morning, says, OK, this is definitely way past the time for organizing classes. But given the special nature of it, we'll try to see what we can do.

And the department is amazingly supportive. So basically within four days, we have the class approved, which may very well be like the fastest anything has ever been approved in terms of classes at Stanford. And then the following week, we have extremely helpful, amazing TAs like Irina and several other students who basically signed up or volunteered for the class.

So that's also made it quite challenging as to how to put all of this together at the very last minute. But I think, you know, with the help of everyone, it's actually been working out quite well. I would say regarding the online nature of the class, I think there are some plus and minuses. So certainly the biggest minus is that we missed the face-to-face interaction with the students, which I really love. I would say the silver lining is that in some senses, it might even be a little bit easier to...

to organize meetings with students in the sense that usually when I teach a regular class, then I just have office hours and people will come in whenever they want to. So sometimes students will come in every week. Other students will wait until the week before the final exam to come in.

But because now this online, instead of having just like a free for office hour, we're actually having an office, like a dedicated hour long meeting for each project group. So there's like a dozen project groups who are meeting with each of them every week. So there's, we have to put in some more structures to really make these online classes work. Right. And I think the structures are, could be quite helpful, especially for these projects based classes. Wow. Okay. Yeah. That's a really interesting, interesting,

So to get back to a point you had a little bit earlier, Irina, you've already had a few lectures that happened in the class, such as overview of COVID-19 and health systems during a pandemic and global response and policies for COVID-19. So based on these lectures, what are some interesting bits of information that you yourself have taken away? I think...

Just the sheer scale of the problem. I mean, I know that the phrase global pandemic makes it obvious that this is a real crisis, but I think what's been reflected in the variety of lectures is that...

This is a problem where there's no easy solution. And so this is why we need experts like our guest lecturers, like people who are in charge of overseeing health systems or people who create and study epidemiological models, doctors on the front lines, as well as people who are thinking not only about state, like local state policy or even national policy, but indeed global response and what policies we could institute for a better future response.

And how about you, Professor Joe? Yeah, very much second to what Irina said. So the other part I think that's really interesting about these lectures is just the real-time nature of it. So we both work in machine learning, and typically for machine learning, our standard benchmark data sets have been around for many years, and people work on that. But here, it's like really you have real problems and real data sets that are coming up, you know,

the day before the class is happening, right? And then we have to try to put this together and organize the students to work on these real problems and the tools could go out immediately. So it's just everything happens much more quickly than standard classes and standard research, which I think is actually really exciting for us and also for the students.

Just to give you a couple of examples, so you mentioned that, yes, we have a class on the policies, global policies, response policies. It's actually given by Professor Michelle Berry, who's a professor here at Stanford, and she's also the director for the Stanford Center for Global Health Policy. So she's really very much hands-on and deeply involved in this. And she's really talking about how do different countries like South Korea, China,

Singapore, China, how are they responding to COVID-19? What policies have they found successful and how does that compare with policies that we're implementing in the US? So it's extremely interesting.

And it very much directly ties into some of the projects, research projects that we have teams of students in the class that are actively working on. So we actually have teams of students working on modeling government policies and doing causal inference to infer the effectiveness of different types of policies like shelter in place.

And the output of that project, we also hope to directly impact the world, different countries and different counties in the U.S. and outside of the U.S. would do later on. So there's this, as you can see, it's very much everything's happening in real time and everyone is working extremely hard on this in real time.

The other example of this is that we also had an amazing lecture from from Kerry Kirsch, who Irina mentioned is actually a frontline emergency room physician in New Jersey. So she gave us examples of how in her emergency rooms. So the patients now have to actually do a lot of their communications through phones.

Right. So because their family cannot come in or even when they communicate with doctors, oftentimes they have to do that through phones, which also presents really interesting challenges. And we have a very much closely related projects in the class where students actually using sound technology.

Data, voice data taken from cell phones, especially coughs that are recorded on cell phones to help with diagnosis of COVID and of other diseases. So there's some evidence, very early evidence, but some reasons to believe that there might be differences in the types of coughs.

due to different types of respiratory diseases. So there's interesting questions of can we actually do some signal processing machine learning to prioritize and diagnose COVID coughs from other types of coughs? So that's very much the kinds of thing that we were not even thinking about a few weeks ago. But because of these discussions with the emergency room doctors and other clinicians, we think, oh, there could be a real opportunity for some new machine learning to have a real impact.

I see. Yeah, the pace at which COVID-19 has hit and really become the new reality has been incredible. And it has also been reflected in AI. And we've been talking a lot about new stories of how there's so many different efforts and projects on diagnosis and so on. So to that point, I'm curious, you mentioned also earlier that the class is very interdisciplinary. There's a lot of different professions and COVID-19 is...

topic where even as an AI researcher, your research really has to be interdisciplinary, right? You have to either think about policy or you have to think about communication or you have to think about how to provide tools to medical professionals.

So do you think this will help the field of AI kind of move forward in being more interdisciplinary and working with other professions and areas on tools and models? Absolutely. I think this actually really presents some very interesting opportunities.

and unique opportunities for our community, for the AI community. Just to give you one other concrete example. So I mentioned that we have groups of students actually working on Twitter analysis. So there's a lot of COVID, as you can imagine, there's a lot of COVID tweets. And one interesting problem is how can you detect tweets

What are the potentially misinformation or false information that are rumors or conspiracies that are being spread around COVID-19 on Twitter? So there's certainly a lot of NLP analysis aspect to it.

But if we want to take it to the next level, right, so you then you want to see, OK, so how does this, once you quantify using NLP tools, like this, the misinformation, various information on Twitter, how do you actually connect that with epidemiological models? Because ultimately you want to see, OK, because of the public sentiment towards COVID-19, that's actually affecting compliance towards structure in place and other policies, which in turn affects the public health measures.

So then there are other groups of students are taking these NLP and mobility outputs and try to incorporate that into epidemiological models. So that's just one example where

really to tackle this problem, we have to consider things from NLP side. We have to look at time series, but we also have to incorporate that into these dynamical systems models. And in the end, the outputs have to be communicated in a way that's really accessible and interpretable to non-experts, to non-technical public. So there's also very much the communications, the HCI components to all of this. And I think that's just really

been extremely exciting for us as researchers. Yeah, yeah. I imagine it must be exciting to at least feel like there's a lot of progress, a lot of drive to develop solutions very quickly. And hopefully if we have another situation like this in the future, we'll be more ready to

To that point of communication, I think it's kind of interesting getting away a bit from COVID that both of you are very proficient in STEM and technical subjects, but also are quite interested in journalism and communication.

So do you think in our increasingly complex times and with AI making so much more progress, we'll need more such cross-disciplinary expertise in technical subjects and journalism and communication? By radio, I should be a perfect person to speak on this point. Well, as someone who is hoping to enter the job market in interdisciplinary journalism soon, I sure hope so.

But I think we're seriously, I mean, yes, unequivocally, yes. I think that the pandemic and a lot of other recent events has shown that there's this imbalance between how much information is being produced, the pace of global lifestyle and trade interaction, and

And also sort of the average human's understanding of how the world works, myself included. And I think that it's going to become very important for people who are fluent in multiple domains to be able to explain things.

and to help others understand that everything in some sense is connected now and that decisions are going to become increasingly more difficult to make given the complexities of our modern world. Yeah, definitely. And yeah, I'm curious, just looking across Twitter and public discourse and all the articles being written,

You've already touched on a little bit, but I guess, what do you take away as far as broader implications for technical complex topics beyond that?

I've been really encouraged to see academics writing more op-eds, for instance. I've been kind of focused on reading a lot about the global supply chain because that's something that I have zero training in and zero expertise in. And so being able to get a glimpse of what goes on in, say, an accounting or a management professor's brain and who's looking at our situation has been very helpful.

I'm going to fan person here for a bit and, you know, shout out to Ed Yong at The Atlantic, who I think is doing a fantastic job helping readers understand that science is not straightforward. These are difficult times and unusual times because the natural pace of science is changing.

is is sort of out the window and and as a result you know people might be getting the impression that science is either broken or not working or or no one knows anything but the reality is that age-old institutions and traditions are being tested um not only in policy and everyday life but also in an area like science where people you know assume that there are answers so you

It's been kind of nice to see across the board good reporting and others who don't usually show up in the communication arena reaching out and trying to get their voices heard. But on the other hand, I guess, to be honest, I also sometimes worry about the amount of misinformation or disinformation that floats around. And hopefully, eventually, the more good information will prevail. Yeah.

Yeah, it's a very strange time where we are all trying to catch up, especially me as someone with very little knowledge in this. I'm basically normal and that I'm trying to educate myself. And I guess hopefully all of us can appreciate science, scientists and journalists more in this time.

And touching on that note of more op-eds, Professor Zhou, what do you kind of take away in terms of having research that touches on health and AI? Would you consider writing more op-eds or communicating to the public more in some ways? Yes, yes. And we would love to do that. And I think actually this class is one effort to reach out to the broader public. And I think actually many of our students

lectures, which we put on YouTube, are quite broadly accessible to the general public. So especially the first few lectures, I don't think you need to be a Stanford PhD student to really get a lot of knowledge and information from those lectures. So that's certainly one effort that we have made to make those accessible to the general public.

And related to the points around science journalism, I think this whole pandemic has also been a really interesting exercise for the entire world to really appreciate the importance of doing good data science and good statistics.

Right. Which is something that's even for machine learning researchers is oftentimes underappreciated when we write our new reps ICML conference papers. And it's actually been really interesting to see the statistical ideas, statistical debates that are happening between that are really happening in real time in public venues.

One very recent one around this is really just trying to figure out what fraction of the general public actually has already been exposed to the virus. That's certainly an extremely important number to figure out if we want to decide when to open things up. And that's very much a statistics question.

And I think a lot of the journalists have been doing a very good job in trying to really discuss the limitations of the existing statistical models and certainty estimates. And I think that's something that's really, I hope to become a really a major part of communications going forward. Yeah. And certainly as someone who runs this effort to communicate more to the public, it does feel like maybe hopefully there'll be more of that.

To the point of these lectures being public, I should mention that the class is CS472. There's actually a public website for it. I was just looking over previously with links to these lectures. So we'll include the link in the description of this episode. Maybe to move on to a slightly less professional note,

Just more generally, now that we are in this shelter-in-place time with the COVID crisis, besides teaching in class, how have you found this period? Are you kind of adjusting to being indoors all the time? Are you looking forward to hopefully soon going out more? Yeah, what do you think, Irina? Yeah, definitely looking forward to going out more. Yeah.

It's kind of funny. I think everyone has been sort of grappling with the tiny little things they miss about their daily lives. For me, I really enjoy cooking. So if I had it my way, I'd be able to cook something different every night from all different cuisines. But I have to plan in advance now, which is fine. Overall, honestly, like...

I feel very lucky that my life has not been impacted very much. As a student, as someone who was doing remote part-time work anyway, and it was doing work, part-time work that is easily transitionable to online. It's honestly been pretty all right for me. I think that, I think that where things got a little bit tricky were as a journalist, as a journalism student and journalist, like,

It's been really hard to keep up with the news, especially when so much of it is

scary when so much of it is uncertain. Um, and, and others and other parts of it are some sometimes infuriating when it's, when it seems, you know, when, when it's not actual news. Um, so it's been interesting to figure that out, like to be in a profession where your job is to, um, keep tabs on developing situations, um,

And I guess I should have known going in that there's going to be an emotional toll in the profession. But that being said, it's been really great to connect with other people in the industry and to be encouraged and to get best practices from them in terms of how to process the news, how to take breaks while we're working. And also to think carefully about how we can best contribute. Because, you know, more news is not necessarily...

more useful for everyone, but rather thinking deeply about like what I can contribute at this time, what would be helpful for other people, and then to do it in such a way that both parties get to remain sane. I see. As a non journalist and someone who has been trying to limit my exposure to news these days, that sounds quite interesting and challenging, but it's good to hear you are learning how to cope.

And Professor Joe, I found it actually my PI, one of the professors in my lab has been telling us that for her, this has been a kind of nice relief from going to many events and staying home and so on. So how have you been finding shelter in place? Right. First, I would say that I think we are many of us are really in a very privileged situation now compared to

what a lot of other people in the U.S. and other countries are going through. I mean, I think around here we are doing relatively well in terms of the crisis and especially being academics, we're in some sense buffered in terms of our everyday lives. So we're certainly very privileged and I'm extremely grateful for all the work that people on the front lines in the service industry are doing. In terms of my personal lives, I think it has been some adjustment periods.

And there are also sort of silver linings in this as well. I, in the past, I probably do not exercise as much as I should during the regular school years. But now because I'm home every day, there's actually no, no, no excuse. At least I couldn't come up with any more excuse to my wife. Why shouldn't go out to run every morning? Right.

So we'll just do some more exercises. So I'm actually running a lot more now than I did before. We're trying to get some runs in every day, which is great. I am very much looking forward to when Stanford and the campus starts to open back up again. Of course, once we account for the safety issues of all the students and staff. And we actually have several...

active COVID-19 projects that we've made a lot of progress on the computational side, right? We really understand what the virus is doing in human cells. And really, we're very much excited to go back to campus and actually test those out experimentally. And to do that, we need to have the actual access to our labs, to the offices, and to the environment where we can actually do those physical experiments. And I'm hoping that that will happen soon.

relatively soon. Yeah, as I think many academics are. And as you both noted, it's good to be aware that we are pretty lucky. I also feel that it's good that we haven't had to change our work life too much. And it's good to keep track of the silver linings, which is something I've been trying to do.

Absolutely. So, yeah, I feel this is a pretty good set of topics. Do you have any things you would like to maybe touch on that we have not talked about? Andre, what's been a silver lining for you? What's been a silver lining for me? I've been doing more things that I haven't done as much before. So I found more time for reading. I've been catching up on my video game backlog. Yeah.

Yeah, I don't know. Just readjusting a little bit and appreciating outside more. Like I've started working on my deck in the sun and appreciating the air, which is kind of weird, but it's kind of nice. That's awesome. Yeah. Yeah, I think that's great. And I think another silver lining is actually perhaps similar to what your PI said is actually these conferences. I actually quite like the idea of having conferences

not all the remote conferences, but at least having some aspect of conferences being remote to reduce the need for travels and make it easier for students to attend, to participate. I think that could actually be a positive development that we can all take away from this even after the pandemic is over. Yeah, I'm quite excited about it. I've been being more accessible and hopefully us having less of a

carbon footprint. Yeah, I just wanted to add to that. That's a really good point, James. I think that in journalism, especially local reporting, it's always like, you know, try to focus on local stories and that's so really important. But for instance, now, if you want to talk to a source who's in like Australia or a different country, or if you want to collaborate on a scientific project, I feel like there are no more barriers because everyone's doing it and it's just the normal way of life. So

I'm pretty encouraged and optimistic about like how, how the availability of video conferencing and meetings for those who have access to it can hopefully help foster more collaboration. Yeah. And I guess we're all learning how to make it work, how to not get too tired from all these video calls and, and be productive at home, which is a useful skill for sure. Very much so. Yes.

And with that, we can conclude this special interview episode of Let's Talk AI. As always, if you enjoyed the show, please rate and view us and be sure to tune in to our future episodes.