Dr. Ming Sun, PhD in Electrical and Computer Engineering | Research Scientist Manager at Meta

In this episode, we discuss how watching Star Wars first sparked Ming’s interest in speech and language technologies, the role of networking and mentorship in supporting his transition from academia to working in industry and his continued growth in the field of language machine learning at Amazon and Meta, and his take on the importance of learning how to zoom out and understand the big picture when approaching research problems.

Hosted by Michael Wilkinson

To connect with Ming and to learn more about his story, visit his page on the PHutures #100AlumniVoices Project website.

Transcript

Michael Wilkinson

Hello, everyone! I'm co-host, Michael Wilkinson, and this is the 100 alumni voices podcast, stories that inspire, where we explore the personal and professional journeys of a diverse group of 100 doctoral alumni from Johns Hopkins University. Today we're joined by Ming Sun. He received his PhD in electrical and computer engineering from Johns Hopkins in 2013 and is currently a research scientist manager at Meta. Ming, welcome to the podcast, great to have you here.

Ming Sun

Yeah, thanks, Michael. It's I’m very excited to be here. Thanks.

Michael Wilkinson

So, I'm also I'm a member of the robotics program here at Hopkins. I also work with bats, which is a weird side thing, but I do a good amount of audio processing, so I'm actually really curious and interested in your work in AI speech. It seems that you've been kind of interested in this field in one form or another since in 2007 when you interned at Daimler. What initially got you interested in that field, and what keeps you interested in that field?

Ming Sun

Yeah. So, you know, I have even like being growing up as a kid, when I was watching those movies like Star Wars I always be inspired by the like C3PO, like, be able to understand him his speech and the translation automatically to me. That's the future of the future of life, and I have been always inspired by in this field, and when I came to Hopkins I learned that there is a I want to say a leading world leading laboratory in the speech language processing as well as in the virtual learning as I find, yes, this is where I am, and that's what I'm really interested in doing.

Michael Wilkinson

So, as someone who's maybe on a little more cutting-edge side of your field with your work at Meta, where do you see your field? You know you talked about like the future, and you know that what the future of speech processing might look like. So where do you see your field going more specifically in the next handful of years? And what are maybe some of the major challenges in the field at the moment?

Ming Sun

Right. I would say that speech at the language technologies has been evolving a lot since the past few decades, you know, from the early days the basic statistics models, and particularly in recent decades, based on deep learning technologies, things has become much better and much smarter. You may see people using those all those virtual assistants, like you could name a few right Siri, Alexa, Google Assistant, people using that on a daily basis to perform lots of tasks. That's where the technology is being used, and I could see that there is still very fast pace technology development. And I would say the next will be to better understand to really better understand what human means, and having those conversations. You probably also see recently, like ChatGPT is a hot topic. That's gonna be one example to make it more intelligent, really understand the semantics and what the people mean, and having conversations.

Michael Wilkinson

So you said the next step is kind of better understanding what people mean like in these conversations, and for the machine learning algorithms to really pick that up. What are maybe some of the challenges in doing that that’s stopping progress from, you know, evolving extremely fast? I mean, I know it's evolving rather fast, but like, what are some of the challenges within that sub space that are like still being worked out?

Ming Sun

Yeah, I would say, it's a it's a very hard problem that to really understand how humans’, for example, brain processes speech or languages and the standard they get the human level intelligence. I think they are still far away to be there. So far we have been getting a lot of great progress, and in terms of using data, a large amount of data to a large amount of models. And I think that's a that's has been proved to be a really powerful approach. That's where that's how it bring us to where we are right now, and what’s next, and how to do things more efficiently. That could be a different path, or how to like make this more environmentally friendly for trying out those models.

Michael Wilkinson

That makes sense. So, you talked about the different paths to solving this. What is kind of the path that you’re on? So how does your work at Meta go toward maybe solving some of these issues, and like in the things that you're working on?

Ming Sun 

Yeah, our team focus on speech technologies and particularly on device, which means we are building smaller and more light footprint small footprint models which we could fit on the devices which have limited resources like CPU and memory. From that perspective the challenges would be how to make the model to be still powerful and robust, within a very limited or constrained environment.

Michael Wilkinson

Interesting. So I know that you did some AI speech processing here at Hopkins in your PhD research, so how much of what you're doing now is relevant to the research you did and like, what are maybe some of the similarities? And what are some of the major differences from your PhD research to like some of the stuff you're working on now?

Ming Sun 

Yeah, I would say that I have always been appreciating what I have learned and what Hopkins taught me and also provide me with a solid foundation proceeding in this field, as well as great like connections and opportunities in the field. So, during my PhD my research was more focused on the general machine learning, which is about analyzing high dimensional data and of using data or information from this pirate domain for your joint interest tasks like crossing goal text document classification, for example. And I was also be taking those courses and research project in speech and language processing which I would say, well-prepared me to start my job, and even today I was always be like like feeling that the my side or the process methodologies I learned at Hopkins still like benefits me on my daily work. So, and I braced down with colleagues about research, about technologies and plans. I would say, yeah, one difference would be at at school, we are mostly working on, like relatively small or moderate size data sets, and we also focus on applications, writing papers. And when I joined the industry one major change I made was to think about like high-level pictures, not only about technologies, also about, for example, why we are building this product, and how that benefit users’ experience and the client, and from there work backwards and define the problem, define metrics, and create a plan how to check it.

Michael Wilkinson

It's really interesting that you know, one of the major differences that you alluded to is this idea of like in your PhD working on much more like focused, smaller scale. And now you're kind of having to go more abstract, large scale, which is quite a different skill set to learn. So, what was that like learning that skill set of how to kind of zoom away from the problem and look at the bigger picture? And you know what some of the things helped you develop that skill set effectively?

Ming Sun 

Yeah, I would say, that takes time and practice. I will say what I learned at Hopkins well prepares me for this. For example, in my school I was being able to get the opportunity to work on the like parallel computing tools, and also all those speech like cutting-edge speech and language technologies, and those things remains, I would say, still relevant, even in today’s context. And yeah, one thing, what would it be that in industry there will be much larger data set, maybe new tools. It takes practice about documentation, about trying and trying and failing fast, and also move fast. So yeah, it's a, yeah. It's a good learning experience. And also, for I would say like getting a good mentor in this field would also be helpful. For example, in the in the companies where I was working at, I was able, lucky to be able to connect to good mentors, and who told me how to run those not only those experiments but how to think bigger about like driving or initiating new project, like driving cross team collaborations, all those with the help to make bigger impacts and really deliver high quality results for users.

Michael Wilkinson

So, you mentioned the mentorship piece of it, which is really important, and I think for a lot of PhD students is always extremely important. What is some very valuable pieces of like mentorship advice that you've gotten over the years? Like, what are some kind of like notable things that your mentors have taught you on a more just general scale over the years?

Ming Sun 

Right. Yeah, there are lots of good advices and coaching I'll have been receiving during the past few years. I would say, for example, during my transition from academic to industry, to be really thinking out of the box, not only about technologies, but also about the product definition and the user experience, working backwards plans, all those would be very beneficial obviously in this environment. Yeah, another one is it's about like how to identify the right team and be able to achieve the right information. As as in those big companies, there are lots of teams and teams will be working on different areas. And typically, it requires the collaboration and the joint efforts from lots of different partners. And there are still being developed on how to navigate in this uncertainty activities and the find the right point of contact to be able to shoot information at the align.

Michael Wilkinson

Yes, that's really interesting, because I imagine you know, as a PhD student, you're never really doing team selection other than maybe selecting like undergrads or something like that. So, I imagine it's a little bit scary, but now that you're in a like when you first transition into a role where you actually are kind of responsible for selecting your team. So what was that like the first time around, you know your first time where you had to actually start building out your team, having just graduated from your PhD? Was that intimidating at all? And you know, how did you navigate that?

Ming Sun 

Yeah, that's a that's a different experience. During my first few years, I was more working on the hands-on work and the very focused, rather focused compared to where I am today. So, from that perspective, yes, I would have still need to have the regular actions with our data team, with engineering team. By that time I was mainly focused on building models and still, when we basically build models in the industry is not only about to get the data using the available data and training, it's the I'll take it as enter ownership that being able to define the data set which is usable and how we collect how we annotate as well as after we train the models, how to integrate into the system which requires a lot of extensive obviously engineering testing. And then to requires lots of collaborations with our data scientist engineers and software engineers.

Michael Wilkinson

So, yeah, so like you said you when you first started, you were working more on the model, like a really heavy modeling side of the team on your team and kind of working with other teams. But at a certain point you had to kind of start selecting your own teams right? So, what was? I guess what was that like when you actually had to go select people for your team now as opposed to being the one kind of like on the ground and working with other teams?

Ming Sun 

Yeah, to me I always say I always think of myself really a bit lucky that it's a rather small transition that I have been the same team working the same team for a while, and then, knowing our colleagues well, and have been collaborating from a long time. So, it's not something new or something suddenly changed. It's more about when the team is growing. And I was also interested in taking out or exploring the management direction, which I think will also help me to team my next step during a career and have bigger impacts. So, I step up as leading a team, and with the existing colleagues better as well as by the coaching team members are doing that. They request a lot of the interviews and extra time.

Michael Wilkinson

Yeah, I can imagine. So, I'll transition a little bit here. You worked at Amazon for almost 9 years, or close to 9 years, post-graduation before switching to your current role at Meta. So, I'll start with, how is it like going from Hopkins to Amazon? How did you find yourself in that role? And what was that kind of like process like of going from Hopkins to Amazon?

Ming Sun 

Yeah, right, so that was, I would say, that was, say a something quite new to me as it was my first full time job. So, I need to make my sides change and adapt to the industry environment. As we talked earlier, changing from focus about academia data set and publication to focus on the product and the user experience as well as industry-level tools and the data.

Michael Wilkinson

So, how did? Yeah, yeah, so I was really interested in talking about like the difference in the kind of skill sets you needed. I'm sure people are curious to know, how did you land that job to begin with? Was there a lot of like? What was the kind of job-hunting methods that you used to get the job at Amazon? Was it just a prior connection you had? Was it something you developers develop from scratch? How did that play out?

Ming Sun 

Yeah, for my specific case, you know, Hopkins has been a worldwide leader in this field speech language processing machine learning so that's naturally a I will say a hiring targets for all those companies. And there are, there are connections, yes, in this field and they sometimes they came to Hopkins and have seminars and talk to the students trying to recruit students from Hopkins. So it was to me that a good that's good opportunity, I think, also, for lots of students at school, and in general I would just say, yes, building connections would be important. That in this field, for example, we have lots of great alumni collaborators in the with Hopkins and with Hopkins and industries, and those who provide good opportunities, and for the either for searching for the internship, full-time job, or even later when you’re thinking about changing job.

Michael Wilkinson

Yeah, I think that's a really good point, especially to your point of you know, a lot of these folks are already coming to Hopkins anyways, cause they're really interested in the students here. So, you talked about the importance of building those connections. How did you actually go about building those connections? Is it just talking to people when they came in? Did you email people, or like, reach out to them on LinkedIn? Like, how did you actually build connections with these various groups?

Ming Sun 

Yeah, I think different people may have different ultimately find different methods to be more effective, right? For me, it's more about actually, I think this as mentioning this particular speech language field, machine learning field, there is a group that that Hopkins has very close connections, and and I certainly benefit from it and appreciate all these opportunities. And even after like working in industry there are also good opportunities. For example, we are doing the modeling or the research side, but we also attend conferences regularly. That's typically a good forum for all the researchers, students, industry partners, or collaborators to come to the same venue, talk to each other, and they say, let’s learn about each other's work? That's it. That will be good opportunities. There’ll also be like recruiting events or other opportunities.

Michael Wilkinson

It makes a lot of sense. So, you went from Hopkins to Amazon, worked at Amazon for close to 9 years, and then you went to Meta. What led to your switching from Amazon to Meta? And like what was that process like after having been at Amazon for so long to then go to Meta? My understanding, you've been there for close to a year now, so definitely shorter amount of time than you were at Amazon. So, what was that like?

Ming Sun 

Right. Yeah. To me, there are lots of similarities between Amazon and Meta that in terms they are like league companies with with lots of good talents, great talent, and large amount of data and financial resources which would certainly help developing technologies in this field. And for my specific case, it's a this opportunity of come out of the networking. Albeit the existing connections that one, and also that after like working at the Alexa for this, like almost 9 years, I was interested in seeing how things are done at a different place. I was very much inspired by the vision of Meta in this, like we are Metaverse field, and I would want to see how I could contribute to it.

Michael Wilkinson

Yeah, that's really interesting. Because I know, you know, for a lot of folks, especially if they’ve been working at a company for a little while, there's a little bit of fear of switching away from that and kind of going into the unknown. But you talk about like this really excitement around it. So, was there any sort of uncertainty for you in kind of the switch? Or was it enough excitement that it didn't concern you as much I suppose?

Ming Sun 

Yeah, it's a, you know, it's a totally new environment. It's a different group of colleagues. That's something new and I need to I need to learn more. In the meanwhile, I was more out, much more excited about these new opportunities. And to me this, yes, there is company changed, there are team change and actually, I was still working on the similar fields as the other. So, what I have learned, and I will say, well prepared me, give me confidence that in this field to continue contributing to it. 

Michael Wilkinson

Interesting. So, because of the work you do, I, in my own personal interests, I'll ask the following question, which is, I know that there's a little bit of debate might be the right word around the way you collect data for machine learning and respect privacy, and maybe some of the controversies around that. So I is that something that you consider a lot in the work you do, of how you do like ethical data collection, how you do ethical data storage? And what are some of the check listed ways that you ensure that you are actually doing that?

Ming Sun 

Right, that's that's a very good point, and certainly become like more and more like high visible right in nowadays. And from my observation, they are all very strict policies at Amazon, at Meta as well, and like we do value an owner user’s privacy and the users’ requirements on project protecting their privacy and the protecting their data. That's to me that's under my observation that's the high priority. Actually, whenever we start with the project, how to process the data, having this, all these mechanisms in place to making sure data is not leaked. It was well protected. And it's also on the users’ request of removing or deleting their data. All those are very comprehensive and very strict policies at those places to ensure that people's privacy and data are highly protected.

Michael Wilkinson

Yeah. That's good. That's good to know. So, you’ve had quite a interesting career thus far. What were your career plans when you started your PhD? Like, did you know that this you were gonna do this to go into the industry? Was there still somewhat of a debate as to what you were going to do? What were your just like kind of general plans when you started?

Ming Sun 

Yeah, that's I kind of know that. Yes, I want to continue with doing research related work, either in the industry or academic. That's how, when I started my PhD. 

Michael Wilkinson

So was there. When was that point where you decided to go the industry route as opposed to doing like a postdoc and going the academic route? And what kind of led to that decision?

Ming Sun 

Yeah, that's towards the completion of my my PhD, and before that, I had opportunity to intern at a couple of places like IBM and IPO. For in this field and I got to close, I would say observation, or the first time experience, seeing how things are done, or research was done in industry. That's quite intriguing that if you talk about machine learning or those artificial intelligence or speech language technologies, those are closely connected to users. You know how to say experience and the use cases. So, from that perspective, there are actually lots of customized technologies we developed at those obviously major industry groups. They typically have lots of great talent and also large amount of data as well as computational resources. So also, they have the. I was also interested me, applying what I learned at school to build things which could be used by a really large number of user base. I find this would be a perfect match.

Michael Wilkinson

Yeah. And I think that is, you know, a very interesting point, especially for us engineers of the industry, unlike maybe a couple of decades ago, is really starting to do more research and you know, like this high-level PhD level research. So, I think you do see more of a trend of engineers going to the industry to pursue the same level of research they'd be doing in academia to a degree, but also in the industry. So, it's very interesting to hear kind of your perspectives on it as well as I hear, you know those perspectives from a lot of folks who are, you know, wanna we wanna build things, right? We wanna build things for a lot of people. So that's really cool. So, on that note, like, what are some of the the fun aspects of your job? And what are some of maybe, like the interesting projects that you're currently working on to the degree that you can actually tell us?

Ming Sun 

Yeah, right. Yeah, I would say, yes, we are working, working out of speed technologies running other device. And as you may think the future, we might have analysis, new devices, or which which you don't hopefully will be lightweighted. 

Michael Wilkinson

I have the insider information on the device. 

Ming Sun 

Right, so, as you may imagine, we could leverage those devices for our daily work or daily communications with our friends, with our families. So that's where we contribute to it. And that's also related to how we think about this VR or AR technologies being used right on those devices.

Michael Wilkinson

Wonderful. So, I got a couple of questions left just because we're running short on time. So I'll- my second to last question I'll ask is, what is you know over your career you've talked about some of the like, the mentorship advice, but in general, what is the most important lesson that you've learned over your career for your own success and your own ability to continue being successful in the things you do?

Ming Sun 

Yeah, I would say there could be a couple of these. One is to be, I would say, getting out of the comfortable zone and focus on things which you think would be making bigger impact. That, for example, by obtaining a PhD at Hopkins, we are certainly experts in a tech field, right? And meanwhile into in the industry to make a bigger impact, I would suggest, think about a bigger picture. How, in terms of the whole system or whole project and why it was built in the first place, and how that benefit user experience and the work backwards from that and allocated resources. And also, yeah, sorry. Another one, yeah, I still would like to emphasize the importance of finding a somewhere like experienced in your team or in innovation who you really trust and think would be giving the good advice and feedback on the your career and the technical development. I find that to be very helpful, and also to be able to ask for help and the support when you make important decisions.

Michael Wilkinson

Absolutely. And it's very interesting that you know, the first one was learning to see the bigger picture, because, in a way, I think that PhD sometimes train you to do the opposite, which is, get very specifically focused on you know the specific sub division of thing that you're looking at and making improvements in that one little subdivision. So it's interesting that in your estimation, a really important piece is to be able to actually like zoom out and see the bigger picture of why are you doing these things and where is the state of the field at right now? And yeah, I think those are all extremely important pieces of advice, for, regardless of what level you're at. I certainly struggle with that. I mean, I know many other PhDs do.

Ming Sun 

Yeah, I would say, it's good to be diving deep, right you know, PhD, and really become experts. That not only give you the obviously, the deep knowledge in this specific field, but also, I feel that I was also trying to be able to identify things and identify problems thinking about how to analyze problems break down into small wise all those methodologies would be very valuable and applicable to many different other problems or tasks. That's that's very helpful. And the meanwhile, maybe it's a first going deep, right? Getting all the inside experience pathologies and then thinking broader about the whole industry and the project.

Michael Wilkinson

Yeah, yeah, absolutely. Well, this has been wonderful. I have one last question for you before we wrap up. What inspires you right now? Whether that's in your work in your just daily life and the things you do. Like what is a source of inspiration for you?

Ming Sun 

Yeah, I was, yeah. You know, it has been almost 10 years since I graduated, and that's also, I think, in the past decade that's also where deep learning took off again, and pretty much transformed lots of fields like machine learning, speech, language, configuration, etc. Or biology, medicine, all those fields. And it's a it's a quite exciting to see that. And it's still ongoing on a very like fast pace. It's quite intriguing to see how new things how often, or how frequently use it came out, and they really change all those how people think about how people's expectation, all those I don't know like what you assisted. So, with all this, this could be quite exciting. And to learn, and the interests, new ideas, and the working on this field and I've also been working in this obviously in this industry, in this field, with the great talents. We really build things which are using by really large user base. That’s that's inspired me on a daily basis.

Michael Wilkinson

Wonderful. Well, thank you so much for taking the time to speak with us, and I look forward to speaking with you again at some point.

Ming Sun 

Yeah, thank you very much Michael again for inviting me and for this conversation.