 It's really a pleasure to invite you to the Celebrating Our Associate Professors event. The event really started in 2018 as a way to celebrate the success of our associate professors who've been through a very rigorous, you know, process to get to the point of getting tenure and becoming associate professors. A very hard process where they had to figure out how things work, you know, find and mentor great grad students, figure out how to get funding, figure out how to, you know, serve so many things that are required for successful faculty and tenure and promotion. So it's really intended in first place as an opportunity to congratulate all of you, but it has many other pieces to it. For example, in the talks that our associate professors are going to present today, they will of course be letting us know about what they're passionate about, whether it's in research, teaching, service, engagement, whatever, but they're also going to probably reflect on what key choices they made that enabled their success, and they'll also be talking a little bit about their future plans, their vision. And that together provides many opportunities in this meeting. Certainly for those who have been advocates and champions for these faculty, it's a great chance for us to celebrate. But this is also an opportunity for all the PhD students, the audience, postdocs in the audience to see, you know, the tangibles and intangibles that go towards success and getting tenure, so that you might be successful in the future as well. And thirdly, it's a really important opportunity for new collaborations. As faculty colleagues join this meeting and listen to what you're doing and how your interests have evolved. This will lead and spark and lead to new connections, new collaborations into the future. With that said, I would like to invite Professor Amy Reidman to introduce our first speaker. Amy is a professor of electrical and computer engineering and she also serves in the capacity of the associate head for faculty mentoring and recognition. Over to you Amy. Thanks, Arvind. Yeah, so it's my delight to be able to introduce today. This is one of our, one of our new associate professors Stanley Chan. Stanley got his PhD and electrical engineering from UCSD, and then got it had a postdoc in both electrical engineering and statistics at Harvard. It happened up until 2014 and in 2014 he was hired here at Purdue in the integrated imaging cluster, and he has a joint appointment currently between ECE and statistics. And in terms of his research he started out looking at image restoration, and then looked a little bit more deeply into signal processing and graph theory. And now here at Purdue he's been looking at computational photography. And his research in single photon imaging and imaging through turbulence. I see that my internet connection is unstable so hopefully you're hearing me okay. He has received a best paper award at our flagship conference on image processing the IEEE International Conference on Image Processing in 2016. He has developed several graduate and undergraduate level data science courses here at Purdue. He's been a strong leader in that educational development of data science at Purdue. And he's received many teaching awards, including the College of Engineering's Exceptional Early Career Teaching Award. The College of Engineering Joel Spira Outstanding Teaching Award, Purdue Teaching for Tomorrow fellow, and I'm really excited to have him. I really enjoy having him as a member of my technical community here at Purdue. And I'm delighted to hear what he has to say for us today now. Thanks. Thank you Amy and thank you everyone for coming. It's my great pleasure to be here to share with you some of the research that I have been doing over the past couple years, as well as teaching experience. The research I do is computational photography. It is at an intersection of three subjects physical world sensors and algorithms. The traditional way of thinking about these three is that you go out to the field and collect some data and then you process the data afterwards. In computational photography, the spirit is that we are going to put all these three together. We're going to rethink about the role of the sensor, how do we put algorithms on the sensor, and how do we put the physical models into the algorithm so that we can do a better imaging task. So in the following couple slides, I'm going to show you some research projects, specifically the two projects that I've been doing for the past couple years. And then I will talk about some teaching experience. Okay, so let's talk about the first project that I'm really happy about it is learning to enable photon-limited imaging with a subtitle of possibly the next generation camera of the CCDN CMOS. Cameras have evolved a lot over the past 100 years, and we're enjoying cameras a lot today. But one problem that is there since the beginning of the camera, which is the low light problem. This is a real image my student captured in our optical room using a CMOS camera in a very low light condition, and you can see that it is extremely noisy. And you may attribute this problem to the imperfectness of a sensor. However, for people who have been working on this problem for a long time, you realize that this is really not the problem of the sensor. Yes, there is some problem with the sensor. But if you have the ideal sensor, meaning that the sensor is perfect, it doesn't have any defect, it has fewer read noise, and this is the image that you would get from the ideal sensor. And this is not a surprise because photons comes as a Poisson process, and that is random by nature. And if you compare this image to what we have used to see in the image processing literature, the comparison is just huge. We realize that. Now you ask what can we do as engineers to do a better job in terms of imaging in this low light condition. There are two things we need to think about. One is can we build a new type of sensor, which I will talk about briefly, the Poisson sensor is a new type of sensor that can that will be a little bit closer to the ideal sensor. Another question would be, suppose now you have a much better sensor, can you do some new imaging classes that was not able to achieve in the past. So, over the past couple years, I spend a lot of time working on a new type of sensor called a quantum sensor. This is a collaboration with colleagues at Dartmouth College, and also a company, Gigajot technology. The new sensor operates in a very special way. As it sees a photon that comes to the sensor, it doesn't acquire the voltage as we used to see, is that it acquires binary patterns. So for every pixel you either see a yes, that is a photon or no, there's no photon. Each pixel is done by a single photon detector. And so you can see radically proof that with this new type of sensor, the signal to noise ratio can be improved, the dynamic range can be improved. But then the typical challenge here would be how do we construct an image from these kind of measurements. So over the past couple years we do quite a few things, including developing algorithms, theory, and so on. A lot of things that we have been able to accomplish and now the sensor is becoming ready to use. Here is a snapshot of the things that we were able to accomplish. The left hand side is a pattern recognition system at extremely low light condition, where we are able to see an image at 0.25 photons per electron per pixel, and that is extremely dark, and we can see that we are able to predict the correct image. The images on the right is a reconstruction that we're able to accomplish, that the fan is rotating is extremely dark, and the conventional approach we either mess up the motion or it cannot be noise. And with our new approach, we are able to do a much better denoising and motion handling. So now let me switch again to talk about another major project that I've been working on, and this is the project of learning to see through atmospheric turbulence. The images here should not be a stranger to many of you. If you have a hard day and you go out and you look at images, these will be the typical turbulent distorted images that you see. To our sponsors in DoD and also other folks, this problem is an extremely valuable problem to them and people have been working on this problem for a long time. So, if you present this problem to a graduate student nowadays, the typical response would be, let's say, let's just grab a lot of data and train a network and get whatever we get. And I've been joking with my graduate students that if you talk to the network for long enough, it will confess it will give you something. But for this turbulent problem, this kind of philosophy doesn't work. The reason is that you don't have ground truth. The turbulence level is unknown, the weather conditions are known, and you need just a lot of these data to train the network. So, we go back and ask, what else can we do in this problem? Can we do something at the root to solve this problem? So, I spent a couple years, almost three years with no publication in this subject. We went through the literature from the 1940s to the 80s, and from the 80s to the year 2000, we went through all these literature, and we always, we need to open up the black box of turbulence physics. So last year, we finally, we are able to develop our first simulator, which is the collapse phase of arbitrary simulator that can mimic that extremely complicated way propagation process. And then earlier this year, we invented a second simulator that is called a phase-to-space transform, and this simulator can speed up by a thousand times compared to this one of the state-of-the-art simulator. So you ask, so what, right? So what, here's the view. If you have a much faster simulator, you can actually generate a lot of images. You can generate thousands to tens of thousands of images in the shopping of time, and therefore you can use them to synthesize training data and train the network. So here is an example using the synthetic data that we are able to generate, and you can get a much better reconstruction compared to the past. During this process, I'm also very happy that we are able to build some equipment setups for you to get these measurements. These are the real measurements, and I'm very happy to do this one-stop shop research in this problem. Okay, so let me finish up my talk by sharing some experience in teaching. I mainly teach machine learning in artificial intelligence at Purdue, and I will say that this is a major initiative at Purdue, and I'm very happy to contribute a little bit to this giant effort. I joined in 2014. The first day I came to Purdue, I started to teach a 300-level course in probability, and I've been teaching it every year. Then in 2015, I created a 600-level course on spot modeling. And that is a PhD-level course. In 2018, I created the machine learning. We had about 300 students in the beginning, and now it becomes more stable. We have 100 students across the entire college taking my course. And in the same year, I created a 200-level course with newly co-carny. It now becomes the Python-related science. I also need to acknowledge a lot of faculty who contributed to develop this course. It's a really really great course now. I also created the machine learning reading group. This is the machine learning reading group at the College of Engineering. This is the weekly seminar. Earlier this year, I created the high school machine learning outreach program that we are reaching to 40 kids per course. And so I can see that some gaps, but I can see that this entire pipeline from high school all the way to PhD-level, I'm pretty pleased to see this complete pipeline during the past couple of years. And if we want to teach data science, we need to have a preview textbook. And this is one of the after that I've been doing. I've been writing this book, Intro to Probability for Data Science, with the goal of you to reduce the textbook price to help students and families. So it will be free PDF and you will be sold at a discounted copy price. And there's some interesting elements of this book, including putting a lot of time thinking about how to write it clearly and explain the different concepts to students. If you're interested, you can go to my website. It's available for preview. If you see any box, let me know. See any typos, let me know. I'm still revising the book. With that I would say thank you. I want to give special thanks to all my students. They do all the hard work. I want to thank my colleagues, also my collaborators, as well as the funding agencies. Thank you very much. Thanks, Stanley. So at this stage, we can open the floor to questions. If anybody has any questions or comments. I have one. Have you given any thought to what what was most influential and steering you in the direction that you chose with respect to how you approach tenure and achieving tenure and where you put your energy as you did so. Sure. What was helpful at this is an excellent question. So I guess everyone has a choice. We have a lot of freedom, we can choose whatever we want to do it and everyone's career success definition is different. I can only speak of myself. I think that's to me I want to be a scholar. I don't want to be a manager. I don't want to run a 30 people map. That's not me. And to become a scholar. I need to focus on a problem that really, that really made me feel excited and over the past couple years is these two problems. And certainly I have much bigger problem in my mind that I need some time to formulate. But the thing that's been driving me most is the research interest. Focus on the thing that you want to do and try to get help from the people that can they can discuss with you brings them with you and work out to hard work and try not to explore the game and just do what you're supposed to do. I think this is the principle that I'm holding. And I think it has been okay with me. I don't know how it would affect other people. It's just me. Okay, thanks. Phil, you have a question. Yeah, hi Stanley, that was really interesting. Thanks a lot. I just had a question about teaching. As a new professor, like, seeing that you built three, I think three new courses, or four, five, I don't know, like a lot. It seems like I've just finished building my first course and it was a lot of work. Do you have any advice like was this too much was this great, like, I was just wondering if you could share some insight on that. I, if I were to go back in time. I think I was just creating one. Yeah, I think it's better to focus on one thing and make it good rather than spreading it to theme. Although I do. I do enjoy teaching this is this is why I'm here. I, I like to write students. However, if I can do it all over. I would say that I would just create my machine learning course to make it the best machine learning course. Yeah, I, I, but I think again, everyone has a different career goal. So I think it is all open to, to, to faculty's choice. We have. We have a question in the chat from AJ he says as a former student of your 595 machine learning class I really enjoyed it. How was your experience teaching students from many different departments. So you're asking excellent question is tough. It's about understanding that background I think I think most of the time struggle of faculty is to to sort of not pay attention to the prerequisites. Not not the prerequisites that we put on the book, but the, the different, the different background that people have. So someone come from a career undergraduate students of course it's different from someone coming from the international school. We don't understand that. And we want to calibrate the students on the first day. What is my expectation, and whether you have the necessary background. If not, here are the resources that I'm willing to help you. If you need more time, don't worry. I'm here to help you don't worry take your time to learn. If you feel that you want to learn in a later semester, don't worry, take it in a later semester. If you feel that you're ready, you're welcome to the class, I will do all my best to help you to learn. You do your part and I do my part and that's what together to to make this very experience wonderful. I think this is how I can, I can help the variety of students. Are anyone have any other questions for Stanley hearing none. I'd like to thank Stanley very much I wish him a lot of success in his future. In his future as an associate professor here at Purdue, and thanks. I just wanted to congratulate Stanley also on his success and, you know, the amazing work he does. I think what stands out is besides he spoke about his research but he's also such a committed and successful teacher and educator. It's great to have you as a colleague Stanley. Thank you.