 It's Muskaan Jindal. I welcome you all here on the behalf of IIT. Today we are here to learn about RBC, TSAI and opportunities for BSC data science and programming students from Professor Bala Raman Ravindran himself. B. Ram Ravindran heads the Robert Bosch Centre for Data Science and Artificial Intelligence at IIT Madras, one of the leading interdisciplinary AI research centres in India. He is the Mindtree Faculty Fellow and Professor in the Department of Computer Science and Engineering at IIT Madras. He has held visiting positions at the Indian Institute of Science, University of Technology, Sydney and Google Research. Currently his research spans the area of geometric deep learning and reinforcement learning. He has published over 100 papers in premier journals and conferences. His work with students has won multiple best paper awards, the most recent being the best application paper at PAK DD 2021. He received his PhD from the University of Massachusetts, Amherst and his master's degree from the Indian Institute of Science, Bangalore. He is a senior member of the Association for Advancement of AI and an ACM Distinguished Member. Now, sir, we would like to hear your thoughts on this. Thank you, Muskan, for that kind introduction. And hello, everyone. I'm just going to start sharing my presentation. I hope people can see. Muskan, can you confirm whether you can see my presentation? Yes, sir, it is visible. I'm just minimizing all the pop-up windows on my screen before I start. So good evening, everyone. And so thank you for coming here to hear about what we do at the Robert Bot Center for Data Science and AI at IIT Madras to start off. So the Bot Center or rather this whole initiative on interdisciplinary data science research at IIT Madras started back in 2014 as the interdisciplinary lab for data science where a group of faculty who are interested in working on something called network science. I'll elaborate on it as we go along. And we kind of got together and started this group back in 2014-13 faculty across six departments. So even back then, it was a very strongly interdisciplinary center. But then 2017, it became the Robert Bot Center for Data Science and AI with support from Robert Bot India. And now it spans 12 different departments at IIT Madras, including departments such as biosciences and humanities and management science, and has 32 faculty members who are associated with the center in some form or the other. And over the last four years or so, we have produced maybe 140 papers. And now we fund from the center about 32 different projects, which are strongly interdisciplinary. It's not just from computer science or any one sub-discipline, but cutting across multiple disciplines. So that's one of the core things I want you to grasp from this talk is that data science, I know that many of you are part of this online data science program, but data science is not something that is just a set of mathematical tools, something that you have to really work with people from a particular domain. So that's essentially why you see me reiterate so much the interdisciplinarity of the center. So looking into more numbers, like I already said, we have 32 faculty, 12 different departments, and we have 80 plus researchers currently, a good fraction of whom are PhD scholars. But we also have multiple opportunities for people to come be associated as research fellows, as interns, and as visiting researchers, and so on and so forth. And that's quite a bit of these opportunities that I'll also highlight towards the end of my talk. And as I said, we have 140 publications and about half of those, about 63 of these are very high impact venues. So more importantly, these numbers are without context for most of you. So I'm going to talk about some of our research areas, the areas that RBCDCI works in, at a very high level, and also talk about some of the application areas that we work in. So you can see that I put on four top level areas that we work in, deep learning, network science, reinforcement learning, and farer interpretable machine learning. So I'll try to give you a sense of each of these as we go along. So let's start off with deep learning. So whenever we say machine is learning, so at the core of it, it's a very simple task that we are trying to perform. So the machine is trying to learn functions from the input to output. So it's given some kind of an input is supposed to produce an output, but it's not told what the form of the function is. If I tell you y equals sine x, then if I give you an x, you can tell me what y is. But instead of that, I'm going to give it to you in the form of these kinds of examples. So in this particular instance, so I'm saying that, okay, if you give me this as the input, then I want you to tell me it is an acclade. If you give me this as an input, I want you to tell me it is a bird, and so on and so forth. So I give you a lot of these examples, and then I'm expecting you to learn a mapping. So there are many ways in which people have been trying to address these problems, especially this kind of working with the image and trying to identify what is that in the image. This kind of image classification problem has proved to be very hard to tackle for machines historically. But people have been trying to work on this problem from at least the 50s. So what happened was that about 10 years back, so people started exploring this kind of connected computing units, which are called neurons. So they're basically looking at this kind of really many, many layered computing units. So then they built together complex architectures such as this convolutional neural network that I put in here. And these essentially operate by extracting more and more apologies. This is actually some problem in getting the slides to change. So for each of these layers, you remember I was showing you these layers here. And each of these layers sorry. So each of these layers start computing progressively more and more complex features that describe the image. So for example, the first layer just looks at patches of color and edges and so on so forth. And then subsequent layers start looking at smaller, more and more complex patterns. And then you can see that higher layers start getting parts of objects, start annotating parts of the object. And then finally at the end, we're able to put everything together and then start learning functions as to how to label these. And is somebody else having control of my slides? Yeah, see now it went back one more. I just pressed only one back. And then I'm not able to control it with my keyboard. I have to use some and I have to click there. I don't know what's happening. But anyway, fine. Let's finish the talk. I like this amount. So this kind of deep learning architectures, allowed people to solve non-trivial problems. Allows people to solve really non-trivial problems, especially this kind of image classification problem. And so back in 2010, people were having almost 30% error on this task. And then when deep learning started kicking in, which is about the year 2014, you can see 2012, you can see that's a huge drop in the error rate. So from making around 25% error, it came down to around 15%. And eventually deep learning architectures started making fewer errors than a human does on identifying objects in any way. So that kind of kick started this whole hoopla. And nowadays people do a lot of research with these kinds of deep architectures, whether it'd be processing images, whether it'd be processing text, audio, other kinds of structured and unstructured data. People have started looking at deep neural networks in a very, very deep way. And so no AI lab in the world can say that they are doing cutting edge research in AI unless they are looking at deep learning. So likewise, in our city side, we do a lot of work in deep learning and related to that, computer vision, NLP, and some of the interesting contributions that we have made are, you know, building an Indian language shoot that my colleague, Dr. Mukesh Capra has done. And so we're doing significant amount of work in this space and not just in, you know, doing just research papers, but, you know, getting our products to the field and working with startups and various companies in order to build solutions based on this kind of deep learning setting. And the next area that I want to talk about is network science. So what do we mean by networks? So networks are nothing but, you know, you can think of them as graphs, right? So graphs that, you know, represent relationships. So if you look at a lot of sources of data in the world, right, so they come in the form of graphs, right? Whether it is, you know, social networks where each node in the graph is a user and, you know, friendships or, you know, follower relationships form the edges in the graph, or it could be like a transportation network. So that's the snapshot of the Indian railway network where each node is a station and the link between two nodes means that there is a track or there's a train running between those stations, right? And like that, you could look at technological networks, information networks, right? And collaboration networks who works with school in the offline setting also. And more importantly, in a lot of biological context, right? Networks arise, whether it is talking about gene-gene interaction or whether they are talking about the structure of a molecule that's biologically active and so on so forth. So people model it in terms of these kinds of graphs and networks, right? And these networks can not just, you know, model one-on-one interaction. I was telling you that, you know, there are two friends, then there's a connection between them. I could also be like a classroom setting, right? So where there are all the students who are taking a single class, right? All everyone of you who's doing linear algebra, right? All of you have some kind of relationship among yourselves, right? So these kinds of higher order relationships also can be captured by special forms of graphs. So in RBC design, we study these graphs and in the various contexts that they appear, right? Not just in the context of these like graphs and computer algorithms, but looking at it in the context of biology, in the context of transportation, in the context of other kinds of interaction, right? So we are primarily focused on using, you know, what are called graph embedding techniques, right? Where you take a network. You remember I was telling you that we learned a lot of features from, so we learned a lot of features from other Bluetooth devices to see if this weird behavior will stop. But anyway, so I was telling you that how deep networks learn features from images, right? And then use it for, you know, identifying what is that in the picture, right? So like that, you can also take networks and learn features from the networks, right? And see how you can use that for, you know, things like identifying who, I mean, what role does a person play or what subject a person likes and so on. So with all kinds of questions that you can ask on this network, right? So that's something there. And there are many, many applications that span social networks, collaboration networks, biological networks, language, right? And co-purchase, what are called co-purchase networks like things on Amazon. So what are things, who bought what together and then you can analyze that and so on, so forth, right? So this kind of network appears all over the place and that's just giving you an example, right? And again, we do a lot of work in this space of networks and you can see that the list of faculty would probably be the largest in this and also they cover multiple departments, CS, biotech, chemical engineering and so on, so forth, electrical engineering, et cetera. And the next area that we work on and something that I have spent a lot of time doing is something called reinforcement learning. So I was telling you, I started off by telling you that deep learning works with large volumes of data, right? And all this, somebody comes and tells you, okay, if this is the input, this is the output you have to produce, right? And then you learn a function, right? Instead of somebody telling you, yeah, y equals sine x, somebody says, okay, if this is the x, this is the y, you have to figure out what is the f that takes from x to y, right? So I told you this is usually what machine learning does and deep learning solves that and you can also answer these questions on graphs or networks, right? But then if you think of how you learn to cycle, right? That was never done from this kind of data alone, right? I can give you hours of YouTube video to watch, you're not going to be able to cycle just by watching those videos. You have to try things, right? It has been trial and error. And so you get some feedback, right? Falling down hurts or there is an adult with you who is going to clap and say, yeah, well done, better or something like that. So that's basically the feedback you're going to get, right? So that is the only source of feedback you get to learn, right? So in some sense, if you look at the image, you know, labeling problem that we spoke about, somebody tells you, okay, if this is the x, this is the y. So if I give you this as x, you have to produce that as y, right? So I'm giving you this kind of instructions. If this is the input, then you have to give me that output. That instruction is given to you, right? But here, in reinforcement, in this kind of learning to cycle, you get only evaluation. So first, I present you with a situation, you have to do something, right? And then you get evaluated for what you did, right? So this, the mathematical abstraction, right, of this kind of a trial and error learning is what is reinforcement learning? So a lot of situations in which reinforcement learning problems appear, right? And then people have looked at it extensively. In fact, I started off with this example of cycling, because many of you might still remember how you learn to cycle, right? Most of you remember how you learn to cycle. But then this is how babies learn to walk, right? This is how babies learn to talk, except it's trial and error, right? Nobody actually tells a baby how to form your local card to produce a particular work. The baby just babbles. And whenever the baby says something that sounds like mama, something, other people just go crazy. And then it gets all the feedback it wants. And eventually, you know, the reward that it gets is that it manages to communicate its need using language. And therefore, it learns to verbalize things, right? So that's how it learns to talk. That's how the babies learn to walk as well, right? So it's a very fundamental mode by which animals and humans learn. And it appears in a variety of situations. A lot of success stories of reinforcement learning is there. Since the olden avatar of reinforcement learning back in the 90s and 2000, so you had things like reinforcement learning that learns to fly a helicopter or reinforcement learning used to control robots, right? Or reinforcement learning used in advertising, computational advertising, or in selecting stories and things for you. A lot of success in modern gameplay, right? Reinforcement learning has a lot of success in modern gameplay. But more importantly, RL also has this recent success in solving very complex problems such as protein folding. It's very hard to figure out what the structure of a protein is, because then only then you know what would be the functionality of a protein, right? It turns out to be a non-trivial problem. And other situations like power control and using, yeah, so using reinforcement learning to actually manage the power consumption in the data center, right? So that minimized reduced power consumption by like 40%. So a lot of situations where you don't know what is the right answer to do, what is the right thing to do, but you can recognize the answer when it happens, right? In such situations, reinforcement learning plays a big role, right? And again, we have one of the largest reinforcement learning groups in the country in RBCT side. And we work extensively with the industry on various application of RL problems. And also we work on building the fundamental RL theory, not just on the applications. So far I've been talking about all the success stories and everything, right? So typically, you know, when all this modern AI works, we are all very happy, right? But then sometimes AI does crazy stuff like that, right? So you can see this, right? So here it was all nice, right? A person riding a motorcycle on a dirt road, great, right? It's labeling things correctly. But sometimes things like this happen, right? So now it says a skateboarder does a trick on a ramp, but obviously somebody riding a cycle and it says a little girl in a pink hat is blowing bubbles, right? So I don't know what happened. So in such cases, I really would like to know why it did that, right? And even in cases where it doesn't do crazy things like this, right? I would like to know why the AI made a certain position, right? So for example, in things like, you know, lending, right? Or trying to figure out somebody should get a loan or not. If AI is saying you shouldn't get a loan, I would like to know why it said that. That's crucial being able to explain or at least being able to interpret what is the answer that the AI gave in such situations. And people actually do things like this, right? Okay, here is the original image and now I tell you there's a dog in the image. Why did you tell me it is a dog, right? And then the AI says, okay, these pixels made me say it's a dog, right? That's basically it. And then how you interpret that to understand whether the AI is doing something good or wrong is still an open question. And this, for the images, it is fine. But now if I start talking about loan decisions, it's becoming more tricky. So there's a lot of work that needs to be done in order to make the AI ready to go out, to face a human who is going to be affected by the decisions of the AI. And that shows up in a variety of different settings. So here is an example of from the Bail Decision Program that I was telling you about, right? So it looked at two people and said that the guy on the left is a low risk offender and the girl on the right is a high risk offender. So don't give her bail because she's likely to commit a crime while the guy on the left is not. But then you go and look at what these people have done, right? So the guy on the left had two armed robberies, attempted armed robbery and after they left him free, he went and robbed, so it was involved in a grand theft, right? Well, the girl on the right had four juvenile misdemeanors, very minor that they have not even recorded. And after she was released, she committed no further offense, even though the AI thought she was at high risk, right? So this notion of fairness again comes into play here. And so obviously the AI is not going to be explained why it did this. If it's going to give an explanation, it's probably going to say the person on the right was African-American, therefore, was at high risk, right? But then this whole idea of AI's decision being fair is still a moot question, right? So a lot of work that is happening in that space, right? And here is another example. So Amazon scraps a secret AI recruiting tool that showed bias against women. So that's again another important challenge that we have to address because women were underrepresented, so they didn't learn how to process women applicants properly, right? So this raises serious ethical questions, right? So AI sending people to jail and getting it wrong, right? And AI thinking everybody is white, she takes a picture of Obama and converts that to a white person, right? And all these kinds of issues arise. So fairness, you know, ethics and interpretability has become a big area of research in AI. And obviously, obviously, there's a lot of work in this. And in fact, we are looking at questions that are very specific for AI ethics in the Indian context. So that's another area that I'm looking at. So I basically outlined four major areas that we work in deep learning, network analytics or network science, reinforcement learning and as well as fairness ethics and interpretability of the AI. So these are more research areas you can come to our web page and know more about the kind of papers that we publish and the projects that are being run into space, right? Then we work on multiple applied problems as well. Actually, we work in four vertical systems. So we look at problems in manufacturing where we work with companies in order to look at, I know, building what are called digital twins. Digital twins or essentially you can think of them as, you know, computer simulations or physical systems that are as, you know, as closely, you know, replicating the behavior of the physical system as possible, right? So quite often people look at the physics behind these systems, maybe it could be a boiler, it could be some engine, you're looking at the physics behind the systems and then trying to build a simulation often. But then when you actually start talking about physical equations, you are making approximations, right? So when you make these approximations and then you build simulations from that, they're going to start deviating from the real system, right? So what we do typically is, you know, use machine learning and AI techniques, data science techniques, in order to correct for those deviations. Instead of building the entire model from scratch, we take models that are built with knowledge of the first principles, right? And then correct these models by adding, you know, AI on top of it, right? And these work much better than trying to build the model completely from scratch. So that's one line of work that we do in manufacturing. And like that, we work with many, many big manufacturing companies in order to solve very hard real problems, right? And the second vertical we work in is financial analytics. And in fact, a lot of collaboration happens with the American Express Lab for data analytics, risk and technology, which is primarily focused on the financial domain and understanding, you know, customer behavior and things like that. So we build, you know, risk models for, say, the Indian population, right? So in fact, jokingly say that somebody comes and offers us collateral to cows, right? Most of the Western models don't know what to do with it, right? So we have to actually look at it from the Indian lens, right? And then figure out what does it really mean, right? And so we are building all these kinds of models, we work extensively with banks and both Indian and foreign banking organizations, and as well as multiple nonprofits in order to work in this space, right? And then the third vertical that we do a lot of work in, in fact, it's become so vast I should probably start talking about it as two different verticals. It's on systems biology and healthcare. And we work closely with IBSC, which is Center for Integrative Biology and Systems Medicine at IAT Metrals. And we look at a lot of what it's called OMIC analysis. In fact, many of the network related things I was telling you about, right? So typically done in collaboration with IBSC on for the biological data. So we can look at gene-gene interactions, we can look at molecular structure, right, and all these kinds of things. So that's one part of it, the systems biology part of it. And we also do a lot of work in the clinical data analysis and trying to build things like, you know, models for estimating the age of the fetus, right? And trying to figure out when the birth would be premature. So that's another interesting problem called Garmini that we are working on has had significant impact and under-crossable, right? So there are a lot of work that we do. In fact, you can come to our page. There are a lot of, in fact, easily accessible blogs written about some of the biosciences work that we do. And so that could, you can read. I mean, you don't even have to understand neither the biology nor the data science to completely appreciate the impact of the work, right? And we also do a lot of work in smart cities. We collaborate with the Center of Excellence in Urban Transportation here in ITM and also with a couple of other centers that work in the smart city space, right? So a large fraction of our work is on smart mobility, traffic modeling, and analysis. But we also look at things like pollution and construction management and water distribution, power grids, managing power grids and looking at connected vehicles and all of these areas also are of interest at the city side. Again, we work with multiple organizations, automotive manufacturing bodies, and the smart city corporation of Chendaya. So that's basically the work that we do in the application verticals. Again, I urge you to look at the center and a lot more space there. And then we look at a whole new research focus that has developed over the past year for the center, which is on looking at something called deployable AI, right? So because we work so extensively in both the fundamental side and as well as working with companies and other people who are interested in taking the AI to the ground, right? So a lot of interesting challenges come up, right? So we have the kind of focusing on research topics that need to be solved before AI can actually go out into the real world and be deployed at various instances. So each of these, when some of these are more general questions and some of these are questions that are very, very specific to the domain in which you are deploying AI, right? So it could be societal challenges, trust, privacy-related challenges, and how do I trust the system like both ethics, partners, business. And there are some data-centric challenges. Where do I get the data from? What do I do if the data quality is not great and so on and so forth? And then there are organizational challenges, right? And as well as actual implementation, like hardware, system-level challenges as well, right? And there are multiple projects that are undergoing, ongoing at AHA BCD site, which are like AI ethics for the Indian context, looking at interpretable models for humans, especially in healthcare, right? And also looking at incorporating some information that you already know about the system. How do I take that into an AI model? So there are many, many projects that we have been looking at. But like I said, we have extensive network of collaborators and you see a sampler here, right? And so this covers both Indian companies, foreign companies, and government organizations, NGOs and all that, right? And of course, our collaboration span the globe. And so, like I was mentioning earlier, not only do we do interesting work, but it's also getting recognized now worldwide in terms of best paper awards at various conferences, right? And some of the top conferences as well from top journals. And we're also having a good social impact. I already mentioned to you about the Garpinny project, right? We are using the Indian data to build a preterm birth categorization, right? But we also work with the 108 ambulance service, emergency response service in Chennai. And we work with some organizations like the Dwarath Trust on predicting low-income families' financial distress, because they don't typically manifest itself in the classical ways. So we have to figure out other surrogate ways in which you can measure financial distress. And we work on with Arman on predicting the risk of expectant mothers dropping out of a healthcare program, like an educational program, right? And likewise, we do a lot of work with other government bodies as well. Again, like I said, as many of these social good projects you can read about in our blogs. And RECDSI not only does research, but also contributes back in terms of tool sets, I'm sorry, tool kits and data sets to the community. And you can look at our GitHub repository for accessing some of these things that we are putting out, right? So there are some data sets on traffic, right? And then the data sets on the language, Indian language, and then there is data sets on various genetic phenomenon, right? As well as some tool kits on, and all of these, like NVDriver and RETEANS also comes with its own APIs and tools. And also some of some other tool kits that are on their way out. And we also work on trying to build standardized data repositories in collaboration with Google and as well as with the government in getting public data accessible to people in a form that is easy to process and do data analytics on, do machine learning on, drive insights on and so on and so forth, right? Very recently, this National Data Analytics Portal actually ran a competition for some selected college kids. So we'll get them to start inviting the online BSE students once we have more people who have completed the diploma as well to participate in this, like looking at whatever data sets are available on the Data Analytics Portal and trying to see, propose different interesting analysis that can be done. So that should be something for you to watch out for in the coming years, right? And we work, like I said, we work extensively with industries. So we have industry consortium where industry partners come and participate in all kinds of activities that happen at our VCT side. And we have this very interesting interdisciplinary dual degree program in data science that we run for IATM students, right? It's a five-year program. So they get a B-Tech in any discipline, but they also get an M-Tech in data science, right? And so they have a lot of opportunities to interface with the industry during this program. And people come from different branches. So we have people with a B-Tech in mechanical engineering and an M-Tech in data science and across the board, right? So there is this interesting program that we run. So once we can talk about a bunch of resources and other programs that we have, that could be of interest to you. So I know many of you have looked at, you know, online courses apart from this BAC program. We also looked at online courses from NPTEL, right? So many of the data science related courses taught by, you know, RBCD side faculty, right? They are available on our web page, but we provide a slightly more detailed interface for them with all kinds of annotations and contents and other things that make it easier to navigate the course. So for example, you can see here a bunch of keywords that have been popped up as content. And you click on any one of this, it will tell you where is the video. That particular video is this word being mentioned, right? And then we are partnering with a company called Video Ken to produce this kind of annotated videos. That's something which you can check out if you are looking for some of these courses, right? And there are many, many, many events, many talks, many workshops that happen at RBCD side, right? And so going forward, we will be, you know, announcing whichever workshops we think are appropriate for the students, whichever event is appropriate for students, we will be announcing these and you can join the live stream of those and listen to these talks, right? When it varies from, you know, industry practitioners talking about how AI is used in the industry to, you know, recent award winners who talk about their fundamental work, okay? And so on. So for then also talking about, you know, somebody mentioned the best application to even be a KDD. So having a presentation on that and spotlight on the various work that we do. And so summer internship program is open, right? And so I think this year might be too early for some of the people who are only doing the online BAC program to apply, but there are students who are, you know, in their third year in undergraduate program or the second year, first year in a post graduate program, we can happily apply to this summer internship. Just check our website for details of this, but I think it's closing very soon. So you should apply quickly, right? And so then we have an innovative thing called the post-baccalaureate fellowship. This is typically for people who have finished their undergrad or finished their master's, right? And so they can explore all, they can be part of RBC DSI and they undertake independent research projects, not like they have to work on something that faculty tells them. And they had some very interesting alumni from this program and almost all of them have gone on to careers in data science, either they've gone on to grad school and doing a PhD in AI in data science or also have gone on to career in data science, right? So again, this program is open now and there are two versions of it when you can apply for the regular post-baccalaureate fellowship, right? But we also have a post-baccalaureate fellowship for women. So applications are open now and people eligible for it do apply. So don't apply if you are in the second year or third year for this program because it's a full-time job, right? And so that's another opportunity for people to work with us. And of course, obviously we have post-baccalaureate fellowship, we hate to have post-baccalaureate fellowships and that we do. And please check our website, I don't think there'll be a great audience here for this, but just to mention that it's there, right? And then there's something that's interesting that we are doing along with the online BSE program, so where we have actually started giving out teaching fellowships for people to join RBC-DCI, do research at RBC-DCI, at the same time also help with teaching become a mentor or an instructor for the online BSE program. So if there are people here who would qualify to this, either it can be as a post-baccalaureate level or it could be at a post-doctoral level. Again, details are all up on our webpage. I encourage you to look at it, right? And so thank you. If you, like I said, right? So we are on social media, we are on the web. So feel free to check us out and we do take questions. Thank you so much for such a great event. It was very insightful and detailed of everything and it also gives us a new platform for new opportunities. So if anyone has any question they can post in the chat box. Sir, there is someone who wants to ask that how can they start their career as a fresher in data science? How can they start their career as a fresher in data science? See, first of all, I'm sort of entirely clear to me at what level you're asking about. I mean, are you asking how do I get started learning data science? Or how do I, I mean, I have learned data science, I have done my online BSE program, what do I do next? It's not clear to me. So if you're asking about, hey, I've done my online BSE program and now what is next? Well, it depends on what you want to do. And the way we have designed the program, right, it's tailored to make you a fully credentialed data science professional at the end of the three years. So you should have no difficulty getting into a data science job. And that is the way that you want to pursue. And the program also gives you the right fundamentals if you are inclined to go for higher studies in data science as well. So if you're interested in looking for a master's again, the program does give you the tools that are needed. So that way, if you're completed the program, I mean, it's up to you. You can choose the pathway you want to go, go as a, go to the industry as a practitioner and then come back and study a little bit more and improve your qualifications. Or just go ahead and go for a higher shields program. On the other hand, if you're asking me, how do I get started in terms of studying to be a data scientist, right? Hey, come on. What kind of skills are expected while applying to the post back program? Is it the same person who was asking me about this? It is from YouTube. Oh, it's from YouTube. Okay, fine. So if you're starting out and you want to know what to do to get into data science, right? One thing is, of course, end all in the online B.C. program. So suppose you don't want to come into that, then there are a lot of material out there, right? So if you already have some good exposure to programming and good exposure to, you know, some of the, you know, the problem settings that people talk about, right? There are ways in which you can start off by looking at these online course material, whether it's MenPetal or other sources, there are enough material out there for you to start learning, right? If you want to become a good data scientist, there are a set of courses that I would recommend that you should do. In fact, like earlier, I used to have to do this to people, right? I had to give a list of courses that they have to take, right? Now you can just go and look at the courses that you need to complete to earn a diploma in data science in the online B.C. program. One of the two diplomas, right? You have one in programming, one in science, right? Look at this list of courses that you have to do to get online, get a diploma in data science. And that, regardless of whether you do it through the online B.C. program or whether you go out and find other material to do it from, those are the minimal set of courses that you need to do to become a good data scientist, right? So that's my answer to that question. So someone is asking, is there any age limit to the internships and all? Is there any age limit to the opportunities of all these opportunities which you mentioned earlier? Age limit to these opportunities. Of course, yes. The, I mean, upper age, right? Yes, yes, upper age. I have to check with that, right? Because many of the students are just out of school. So you're asking about an upper age or lower age. That is an upper age limit, especially for the post-baccalaureate program. I think the upper limit is 27 years as of this June or something as of the coming June. You shouldn't be more than 27. And it's, I think, 28 or 29, if you apply for the women program. And for all of the programs, they're open to both men and women, except the women post-baccalaureate program. The only difference between the regular post-baccalaureate program and the women post-baccalaureate program is actually, because women are typically are fighting a lonely battle in technology. So what we do with the women post-baccalaureate program is that we, we arrange for mentors, both from the industry and from academia, women mentors, to come and meet with these post-baccalaureate fellows and have conversations with them and also answer their questions about surviving as a woman in the technology field and things like that. So that's kind of a little bit of an additional assistance, right? In terms of being a woman and making it in the technology field, right? So that is one of the reasons we have a special fellowship for women, nothing else. Otherwise, everything else, as far as the benefits, the qualification norms, we apply and all of these are the same, right? And so the age limit is there. If you ask about engineering versus non-engineering, I saw one query about somebody who's doing a BSc honors in mathematics, right? And that's fine, you can apply, right? So you can apply even with the math background. So the only challenge is that for the internships and for the fellowships, right, we do give you a coding challenge. You have to get through the programming challenge before you can, before we will interview. So if your programming skills are very poor, then you won't be able to make it through the coding challenge. So even if you have a math background, getting through that is important. So it's not like super fancy computer programming or anything, but you should know how to get through that, right? So for three-year degrees is a little bit of a challenge for the online BSc students, right? So we are going to open up the internships for anybody who has completed the first two diplomas, right? So that would be typically mean you would have done two years in the program, right? But since it's our own program and we know what we are teaching, so we are opening that up, right? If it's a two-year degree elsewhere, we will have to take a call, right? So you could write to us, we might ask you details about your, for example, CMI is three-year degree, we're happy, right? We're happy to take people for internships from there. If it's a three-year degree from elsewhere, you'll have to apply, write to us and check with us, right? So I wouldn't close it down, but I also won't throw it open. So we have to check with us. Yes. And you have to tell me, I mean, the questions have just been scrolling up too fast, I can't keep up with myself. So I mean, a few interesting questions to make. Okay, someone is asking what kind of skill sets they can expect when applying for post-backup program? Post-backup program? Post-back program, post-backup. What kind of skill sets do you expect? What do you mean by that? So, okay, so we ask people to, okay, I'll tell you what the selection process is, and then you take your call, right? So the first thing is, we expect you to write a meaningful statement of purpose, right? So you need to write something that's just like you're applying for, you know, graduate school abroad, right? So you write statement of purpose, which kind of outlines why is it that you want to do this program? Why is it you want to do research? And, you know, is that any particular faculty or a particular project in the center that interests you, or if you have done any interesting projects in the past, right? So we take the statement of purpose very seriously, right? And in fact, we have relaxed, people who have written stellar statement of purpose, we have relaxed even the marks requirements for them to get it, right? So that's one thing. The second thing is, we look at the coding challenge, right? So you have to go through a programming round. And if you do well in the programming round, only then you progress to the next round. So after the programming round is done, the next round is a technical interview. So we ask you questions on the fundamentals of machine learning and things like, you know, can you do understand regression and very simple questions, right? So nothing that a person who has gone through the online basic program won't taste, right? So it's easy enough. But these are like basic fundamental questions in machine learning. So and only then after that, you actually have an interview with the faculty member, right? So this is, so the interns don't interns basically do the programming challenge and have a very minimal, you know, technical interview, and then we select the interns. But with the post-partner fellows, we have the statement of purpose is the first screening and the next the programming and next you have the technical interview. And then you have an interview with the set of faculty members. And you choose which faculty member you want to apply to, right? Who you would like to be a mentor for you in the post-partner program. And then you basically get interviewed by the set of faculty. I mean, you can choose one, two, three, how many you want, and then you get interviewed by them. And periodically, we put out, you know, special calls, right? For example, we had a large COVID data analytics project where we put out a call for post-partner fellows in the COVID to join the COVID project, right? So that, in such a case, then you might get interviewed by another panel that's working on the COVID project, right? So, but if you're looking, asking skill sets, programming and fundamentals of machine learning are two things, important things. Okay, that's good. Okay. Yes, there is a upper age limit in the applying. So what can- The upper age limit is for the post-partner fellowships and the internships, right? Okay. So, post-doc, post-doc, post-doc, the upper age limit is sadly wide, I think. Maybe up to 30 or something. Some people are asking that there are people who say that there are no jobs for fresher in data science. Can you say something about that? Oh man. That is a hard question to answer. I'm not sure. See, there are too many programs out there, right? That claim that they are producing data scientists. In fact, they don't give a good background in math. They basically teach you like one or two tools, how to use those tools to solve regression problems, and they say, oh, we teach you how to identify people from photographs or something like that. And there's like just like one simple setting of a tool, and then you give it a data set, you run it and the output comes something. That's not really training you to be a data scientist, right? This basically gives you some kind of minimal hacking skills in putting together a data science solution. It doesn't really make you a data scientist, right? And so what you really need is a program like this online basic program or a more rigorous training like we provide with our data science masters, right? Where you get a good grounding in the fundamentals of math and you're able to understand and interpret domain-specific problems in the context of data science, right? So what is the data science issue here and so on and so forth? So once you get this kind of a grounded data science profile, those are the kind of people the industry is looking for, right? Industries now has its fill of these data hackers, right? I mean, the people who can just run a little bit on, you know, know how to do three tools, right? And then if you tell them what is a machine learning problem you want to solve, they can solve classification, they can solve regression and basically that's it, right? Those kinds of skill sets, right? I think that those are saturating the industry and people are not really interested. So if you want to be a data scientist and look for jobs, make sure that you are a grounded data scientist. So if you're going out of this online basic program, I think you have not much to worry about. But yeah, the industry landscape is the fast evolving one. But I think if you have the right fundamentals, you should be fine. Thank you, sir. That will help us to increase our career. Our next question is after learning about the algorithms of machine learning, what kind of projects or research topics should we try to address as a beginner? Oh, go ahead. So there are many interesting problems that are out there. But if you want to do research, right? Just after you understand the fundamentals of data science, maybe doing the first year course or something like that, it's challenging for you to do it independently. You really need to have some kind of a mentor, right? And so I would strongly recommend after you finish both your programming and your data science diplomas to strongly look for internships, whether it is at IT Metas itself or internships elsewhere. So we are talking to a lot of other partner institutes who are happy to open up internships for online basic students. So hopefully there will be a lot more opportunities. So I would strongly recommend doing this kind of a research internship, right? That will give you the exposure you need. And if you try to independently get into research after just running the fundamentals of ML or fundamentals of data science, it's going to be a hard slot. There is no one-size-fits-all answer to these kinds of questions. So if I give you this kind of generalist answers, don't be disappointed, but that's the truth. You have to look for an internship. Yes. Sir, some people are asking that there is an option of give ranked, but they don't have any rank. So can they apply? Is that optional? Of giving your rank in the form of internship, summer internship. If you don't have a rank, it's fine. If you have a rank, it's optional. You don't have to fill it in. The internships at RBCD, I just one question just popped up, how long are the internships at RBCD inside? They vary. We have internships that run for eight weeks and we have things that run for six months also. I mean, you have to look at, so for our internships are not just you apply for an internship. There are projects that are listed on the webpage. You apply to a specific project. So each project will have varying durations of internships. So you can apply to those based on that. But you have to, if you're only in the online BAC program, you need to have two diplomas before you can apply. And if you're also doing an additional program elsewhere, you can apply if you're finished three years. Okay, some people are asking that streaming round in programming exam, there is it related to machine learning or problem solving, like, which is it? The programming exam, it's machine learning. It's very, very, very fundamental ML things. You can easily crack it. I mean, you don't have to be an expert in ML because all right. So I'd like to answer this one question. I'm a chartered accountant with quality 20 plus years experience and so forth. So basically the question is, are there options for post graduates in data science and RBCD side? So the post-baccalaureate fellowships, ironically, even though it's named post-baccalaureate fellowships is also open for people with a master's degree. But that is an age limit to it. If you have 20 plus experience, it's going to be a little challenging. Where did you write to me, KV? And then we can figure it out. The internship is currently planned to be offline, but we are happy to consider online mode as well. And if there are something that prevents you from traveling offline, but the only condition is that you should come at time to the internship. You can't be doing something else and then say that I'll do the internship offline. So we do have some kind of a minimum bar on GPE. I will not tell you what it is. But if you write a good SOP, then maybe we will make that. So let us have recommendations are not mandatory. Somebody is asking, do I need a letter of recommendation? Let us have recommendation are not mandatory for the post-baccalaureate programs. Only for the post-doctoral fellowship, let us have recommendations mandatory. For the other programs, we don't need it. So applying for an MRs at IIT Madras after three years, right now, no. But that might change it on the right. I don't know. But right now, regulations say you need to have at least a 48 degree before applying for an MS or you should have done a BSE, MSC somewhere and then you can apply for an MS in an engineering discipline. The questions are not interesting that I've asked so many questions. So the rank in your college and letter of recommendation should be optional. So somebody can very quickly try it out on the web page and tell me if it is mandatory, then I'll change the form. Okay, somebody is asking me the lack of coding and software tooling is the biggest assembling block in AI in India, we agree. A lack of AI skills also. But all of it, all together system development skills are poor in India. That's one of the biggest challenges. We're hoping to address this through this program. So somebody is asking, is this open source contribution mandatory list? Nothing is mandatory. It is mandatory. What is mandatory? Oh, LOR is mandatory? For what program are you applying? It is open source contribution mandatory. It's just asking that. It's not. It's not. Open source contribution is not mandatory. And so rank is mandatory there in the form. Rank is mandatory in the internship form. Okay, I'll ask them to change it. But if you're only doing the online VSE program, you cannot apply for the internship until you finish both your first two diplomas, right? Both the programming and the data science diploma. So it's currently, I don't think anyone has finished both. And so you can't apply. I think so. That's fine. Ah, okay. So rank is mandatory. I'll, I'll, sorry about that. But tomorrow, try again over the weekend. We'll remove the rank. And that's all sir. Thank you so much. For my email address, I just put that in the chat window here. Okay. For people on YouTube, it's Ravi at cse.itm.ac. I mean, you can find me by Googling me here. Somebody is asking, can I play for multiple internships? You can't do that concurrently. But if you're applying one after the other, sure. Okay, fine. I think we should stop. A lot of interesting questions. And is there any way that you can collect all this question to, and then, and then I can try and answer them offline. And if there is a way, you can share it to the other students. Yes, we'll try. Thank you guys. And I'm happy to see so many people actively asking questions and participating. Thank you so much, sir. It has been such a great event from your experience. Thank you so much. Thank you guys. Take care. Bye. Bye sir.