Sure.

Just a quick intro, Parvathy — can you tell me about your professional journey in academia so far?

Hi, I'm Parvati Rama. And I come from Kerala, India. I've recently completed my PhD in Mathematics from Amrita Vishwavidyapedam. Amritpuri Campus. In Kerala, India. Uh, I have studied. Uh, pancreatic cancer. That I have incorporated. Mathematical Modeling. Artificial intelligence. Healthcare and image processing into my work. I have studied the tumor. From the perspective of diagnostic. Uh, prognostic. And uh. Uh, in all these manners. I have produced uh but. Uh, some papers and UH-4 patents from this. I have developed. Uh, a couple of. Hybrid mathematical AI models. And also. Uh, decision-based models, which is actually deep learning based. For this particular model. Uh, when it comes to my other educational background. I have completed a five year integrated MSC course. Which comprises of both buying masks. Vectors and bachelor's degree. From the same campus, Amritapuri Campus. In the year 2020. Talking about my long term goal. I'd like to work in a prestigious institute. Especially in an dictionary. Department where I can use both my mathematical and technical knowledge and also bring. A balance between being a full time teacher and a researcher in academia. So that's why I chose vit. So that's everything about myself. Thank you.

You mentioned combining mathematical modeling with AI and image processing in your research. When you're teaching a new class, how do you introduce complex concepts like these to students who might not have a strong technical background yet?

So my idea of teaching us. Everything can be understood. If we. Make it into the application side. So for example an image. It can be seen as a metrics where each pixel is a. Uh, numbered A numerical value. Where it represents the intensity value of the. Pixel. So if we represent this image like. Everything that we see around us in a mathematical way, especially in an intuitive manner. Students might understand it. So image processing is basically just playing out. Uh, with these matrices? And artificial intelligence. Basically, it's mathematics. So we can I'll explain students how. Each of these things can be. Worked out using mathematical formulas or equations and how it can be applied to real-world problems that they see. For example, if. They have some. CT scan with them. Uh, which is grainy and they want to see, uh, their. Uh, any if there is any abnormality? In some so I'll make them. Process the preprocess the image. And I'll. Teach them how to use mathematical equations to work on it and see if there there is any abnormality. So uh, if they are applying it in with an intuitive. Manner I. Believe that anyone, even if a student who is not very interested in the subject, will be. A curious to know how? The working is so that's my way of. A teaching, I believe.

Okay, so you try to make things intuitive by showing how images are just numbers and mathematics is behind the scenes of AI. Can you give me a specific example of a classroom activity or assignment you've designed that helps students grasp a difficult concept in this way?

Sure. Uh, so I'll make them bring an old photograph from their family photo where it is grainy or. It has some noise in it. Uh, and I'll make them upload it into their system. Where they can see their image and I'll make them understand that each of these pixels are numbers, so I can use either Python or Matlab for that. And each of these pixel values can be represented as. Uh, numbers between 0 to. 256. And then I'll let them know that there will be some. Uh, numbers, which are noises. Which might not. Follow this. Follow the. Uh, numbers of other. Pixel or neighboring pixels? So I'll make them understand that these are noises and we need to remove it. For their old photographs to be. Uh, seen as new. So, uh, I'll. Make them or uh. Help them for denoising it. Using any algorithm? Uh, so that they can see or they can do it in the lab. Such that they themselves make their own photographs new. So that is 1 concept. That I'll be, uh, making them do. In living. Classroom.

Alright, in your recent publication in Cancer Treatment and Research Communications — the one on attention-guided residual 3D U-Net for multi-modal pancreatic tumor segmentation — how do you translate the core ideas or methods from that research into something accessible for your undergraduate students?

So in that paper what I did is. I have taken UH CT scan images of pancreatic cancer of an abdomen and their pancreatic cancer is present, especially pancreatic ductal adenocarcinoma and they've used. A modified unit model which I've named Catnet Unit. Check unit 3D. Where, uh, it segments, uh, this particular tumor from the CT scan. So there simply I have done. Uh, the segmentation part where? Either uh. I have I had training data where? The, uh, everything else other than the tumor is. Marked as background. And the tumor is marked as. Uh #1 so everything else is 0 and this thing is 1. So what I did is I have predicted everything. Else is 0. And my tumor S1. And trained my. Model according to that. So there I've used. Mechanism where? Special attention is given to infiltrate tumors. That means. Some tumors might be very difficult to be seen, especially if you're using. Uh, computer vision. So there, uh, the boundaries are given special attention using. From some mathematical formulas. So that's how I have translated the entire. Uh uh thing into mathematical side. And then segmented it.

Alright, so you marked everything except the tumor as zero and then used special attention for tricky boundaries in your segmentation. If a student struggled with the idea of attention mechanisms in deep learning, how would you break that down for them in a way that clicks?

So I believe uh, the cities can and segmentation of uh tumor is bit. Tricky I. Agree with that. So I think I'll start with small images where, uh. Background is not very loud or noisy. I'll start with an image of fruits where only. A tomato is. Red of color and everything else is some other dull colors. So I'll make them understand through that. How everything else is 0? And the red color is. As marked as one. And I'll make them do the mathematical formulas there, where this attention mechanism is applied into only the boundaries of the tomato where they can see what is happening. Every other colour I'll be giving some, uh, filter where the colours. Less than uh, this range should be marked as zero and greater than this as one. Where, uh, they can actually see and replicate. What is being done? So they will understand the intuition behind this particular idea. Then I'll be moving on to some more complex images. And by step by step, they'll be understanding it. And then find. Finally, the CT scan because in the case of CT scan. Uh, normal, uh or nor, uh, unprofessional cannot I understand what is being done there because it is very complex. Not only a healthcare worker can understand it. So I believe if I start from small images. And build the knowledge step by step. The student might be able to. Uh, do it, uh, precisely, I believe.

Alright, Parvathy, I’d like to focus for a moment on how you’d innovate in the classroom. Suppose you’re assigned a large undergraduate theory course—let’s say 180 students—and slides aren’t allowed. How would you design a session that keeps everyone actively engaged and ensures they’re genuinely grappling with the material?

OK. So if there's a large number of students, say 180 students, that that means it is a large crowd. So if I'm just teaching on the board, uh, I'm not sure whether the students will be understanding, especially the students sitting in the back will be understanding. So what I will do is I'll divide the classroom into, say, six groups. If I say so, there will be 30 students each. And I'll make each of them. I'll give each of them assignments I'll first. Teacher concept, for example if I'm teaching linear algebra. I'll be teaching about. Uh, vector spaces. And I'll be giving each of these groups separate assignments so that, uh, all 30 of them should contribute to it and give me an output or an answer. After a few minutes or after an interval that I give them. So that. Every group can ask me questions. Individually, if it is a whole classroom, it is very difficult to answer 180 students at a time. But if I am making them groups and asking them questions or if they are asking me questions individually. As a group of 30 students, it will be easier for me to handle them. And it will be easier for them to. Uh, have a conversations with me. So that is, uh, one thing I'll do also, I'll give them group projects, just like I said. Uh, and. I'll be giving uh special classes for. Students who are not. Uh, getting the concept properly? So I'll be doing test and. Marking them is. Some students are below par. I'll be having a group discussion with them after the class or something like that. So that is how I'll be. Uh, uh. That that's how I'll be dealing with this problem.

Since you mentioned running group projects and managing large classes, let me ask about institutional service. Beyond teaching and research, what role do you see yourself playing in department-level governance — things like curriculum committees, program reviews, or accreditation efforts?

If I'm given a chance, I'm, uh, willing to. Make curriculum changes so my idea of. Teaching is especially to engineering students. They should more be. Uh, looking forward to the. Industry side or the application side? So if given a chance, I'll be. Uh, developing curriculums. Based on students interest, the industry needs and the application side. That's one thing, uh, I think I can do, I'm confident in doing. Uh. Then uh. I can. Uh, maybe suggest a day for? Uh, the workshops or webinars where, uh, I'm willing to organize. Uh, useful workshops or webinars for students? Uh, which? Which might include had sound sessions. Uh, where students? Are interested and might. Not be getting enough. Uh, attention to. So that's. Uh, second thing, uh, I'll be interested in doing in the department level.

Alright, you mentioned you'd be confident participating in curriculum development and organizing workshops. Let's say your department is in the middle of an accreditation cycle, and you notice that outcome-based assessment data from lab courses isn't being collected consistently. How would you address that issue day-to-day?

So that is 1 real problem I believe. Uh, what I would suggest is after every lab section, students are given around 20 minutes. Time to. Uh, analyse them or given them a? Uh, problem. A real time problem so that they can do it and. Submit it. Before a certain time. So suppose uh. They have a lab section for two hours. The classes will be done. They think the first. Uh, some UH-1? 100 minutes and then. The rest of the 10 minutes will be given for them. Address an issue or one problem that might be given, or one objective that is given to them. They need to do it. And sub. Uh, that day itself. Or, uh, within the midnight of that day itself. Is that every week? We can collect. Uh, the. Worksheets of the students. So that when it comes to the, uh. Uh accreditation. Uh, part of the year. Uh, all this, uh. Worksheets are with us and we can evaluate it and put a grade to it. So that is one solution I am thinking of for this problem.

Alright, Parvathy, let me give you a scenario. Imagine a student submits a formal complaint to the dean, claiming your grading was biased on a major project. At the same time, your department head is urging you to improve pass rates this semester. How would you handle the situation, balancing academic integrity with the pressure from your department?

First of all, I I'll. UH can talk to the student and I'd like to know why he or she might. Feel that my system was biased. Uh, I'll make sure that that thing never happens again. So, uh, my grading system, what I'm planning is I'll be keeping notes on how I grade the students. How the grades are worked, The answer sheets will be there, Each of the questions, how many marks are given will be recorded. Everything will be there. So that I can prove that my grading system. Never biased because I think as a teacher it is never OK to be biased in front of the students. So I talk to student individually. And I'll be uh. Discussing how or why he or she felt like that. And I'll handle that situation like that. When it comes to my department head. And that. Uh, person needs my. Uh, great to be improved. So one thing I can do is I can. Actively make changes in my class I'll. Make my students work more. If possible if they are lacking behind if that is the scenario. I'll make them work more, I'll give them more assignments and worksheets and will make them. Score mode. If that is not the case, my students are working the best but. My. Class average is less. Which is definitely not. Fault of my students. Then I might talk to my department head the. What is happening my students, maybe because they they are struggling with the basics, maybe in the previous semesters they were not really. Thought what are the? Basics and stuff. So that's might be the reason they are lagging behind. So I'll talk to my HOD. Uh, regarding this and I'll let them know that I can't illegally give marks to students just to make the class average. Higher, but I'll make everything in my power. Possible to make the students actually get good marks? So. If such a situation comes, that's how I'll be dealing it.

Alright, switching gears now — given your background in mathematical modeling, AI, and image processing, do you have connections with industry or healthcare organizations that could lead to internships or real-world project opportunities for students at VIT?

So I have done my research in Amrita and have collected. My. Uh, data from uh, Amrita, uh. Hospital in Kochi. In the Institute of Medical Science, Kochi. There I have an active collaboration. Uh, from which my latest publication? OK, so I think I still have the connections and if my students from VIT are interested and are intuitive enough to work in healthcare. And mathematical modeling and AI. I definitely believe that there is an option to collaborate with the doctors from AIIMS Kochi.

Alright, let’s shift to a few quick checks on core skills. You’ve worked at the intersection of AI and multimedia in healthcare—can you describe a situation where you had to choose between a classical image processing technique and a deep learning approach for a real problem?

Yeah, sure. So when I started, my intention was to just stick on to the mathematical modeling part. Where I can study the tumor in different aspects? But while I was studying using a classical mathematical models or hybrid mathematical models, uh, there was a lot of data in my system. So my work. Contains around very large data. And it was. Very difficult to just. Uh, solve everything manually. Or using any other computer programming. Systems. So that's when I thought of using artificial intelligence to it. And I started with the gradient based models. For my first initial. Uh, AI? Hybrid models. But in that case also the data was very huge and my model was in performing well. So that's when I thought of deep learning models. Where it can handle actually a large set of data and the noise was also not a problem in that. Also when it comes to images, there will be so much of data in it. And each of these pixels acts as a data point where manually doing it. Feels, uh, kind of impossible. So that's when I decided that I'll be going towards the deep learning. Path that I can. Uh, develop decision-making models. Uh, using it? So that's how I. Uh, gone to the decision, uh, the deep learning. Model path. From my classical. Programming.

Alright, we're wrapping up. Is there anything you'd like to ask about the role or the team?

I'd like to know uh, how my, uh, role as a teacher and a researcher in vit. B as a fresher and I this is my first institute I'm going to work on so I'm curious how it will be. NVIDIA.