Good day, Dr. C. Kavitha. I'm Arjun. I've had a look through your background and your professional record. I'm interested in hearing more about your work and how you see yourself fitting in here with our faculty. Ready to start?

Please start by introducing your professional journey in academia.

Yeah, sure, myself, Doctor Kavitha. Currently, I'm working as associate professor and head of department in Electronics and Communication Engineering, Srinivasa Institute of Technology and Management Studies at Jittoor. I have around 13 years of experience in research, teaching and academic administration work. I have completed my PhD in BAT Vellu in the year 2017 in Machine vision system.

I completed my Mtech from Government Institute of Technology WILLU with a percentage with the CGP of 8.2 and Btech in Electronics and Communication Engineering with an aggregate of with a percentage of 81 in Salem Solar College of Technology. Recently I have completed a city sponsored 6 months QAP program Quality Improvement Program. On machine learning and deep learning.

In an at IIT Trichy, sorry AAA Trichy and completed with the percentage of the CGP of 9.33.

My uh, research work started in digital image processing, uh, including machine, machine system and fuzzy logic system. Later over the year, I have shifted gradually towards applied artificial intelligence and machine learning for real-time applications engineering. Real-time engineering problem next, uh, currently I am focusing mainly on, uh, electric vehicle systems here. Uh, I work on battery based data analytics and.

Uh, modeling of.

Thank you for sharing that, Professor Kavitha. It's helpful to understand your background and focus areas.

Thank you, Sir.

Yeah, sure, Sir.

You're welcome, Professor. Shall we proceed with the evaluation?

Let us start with your expertise in Artificial Intelligence, Machine Learning, and Data Science. Could you describe one of your research projects in this area, emphasizing the methodologies and algorithms you used?

Yeah, sure, Sir. Recently I have completed one project that is SoC state of charge estimation in electric vehicle. So generally the what electric hybrid electric vehicle heavily rely on lithium batteries and the battery will degrade over the period. So in that the very important parameter is state of charge. So this estimation is very, very important to analyze the performance of the battery.

So here already some conventional method is there like uh, Coulomb counting method that is uh current based method, voltage based method under model based method is also there that like Kalman filtering, extended Kalman filtering all are there but it has some limitations like uh, if we take a conventional method, current based voltage-based method. So there the error will accumulate and it need a long rest for measuring the performance of the battery. Whereas in case of. Model based it will work well, but we that.

The assumption there was the battery was considered as a stable one. But in real cases the battery behavior is a nonlinear one. That nonlinearity was not included in the existing method. So what we done was we we have included data-driven based model by using a machine learning algorithm because the machine learning algorithm can work better even for nonlinear cases also. So since here battery is a nonlinear one, its behavior is a nonlinear.

1 So I have included here machine learning based algorithm. So here 4 different machine learning algorithm was was used in that, uh, a real time data was taken in the real time data. It's a real time data means here it's a sensor data. Only the sensor's data like battery temperature, voltage, current, ambient temperature, motor temperature, uh, it's speed, everything was measured. And that is collected and preprocessing was done. So pre processing here is if there is any missing value.

Do if there is any outlier that should be removed because for a machine learning we have to give a structured data. So here to make a data as a raw data as a structured 1 the pre processing step was done and the outlier missing values, duplicated values everything was removed there. Next is feature selection. So feature selection here in our data set we had 96 features. And all 96 features will not be useful for estimating the SoC value, that is state of charge value of the battery.

So in that some feature selection method was used and here the method used was MRMR method that is maximum redundancy. Sorry minimum redundancy maximum relevance method was used. Why this was used means suppose if. Here we have checked the independent variable that is input as well as the output relation. If that is highly correlated means that data that feature was included in our data set. Suppose if.

Two datas are not related. Not relevant means the data was removed in this way. Out of 96.9 parameters, 9 input features were was selected. In this method. These nine input features were was again given to the machine learning algorithm. Here the used machine learning algorithm is Support Vector Machine, Neural Network, Random forest and Gaussian process. Gaussian process regression. So all here is a regression one because. SoC here is a regression value and it's a numerical value only and it is continuously varying. So we had used.

Regression based model here in that we have checked the performance of these four models and tested this also. Out of these 4 Gaussian based regression model gives better accuracy and these models were evaluated by using some evaluation matrices like root mean, square error. Mean square error map mean absolute error error. So these me sorry me these ****. We have used for for evaluating the performance of model. So out of these 4 model. Gaussian based model works better because Gaussian based model depends on the probability so.

Probability can be applicable even for uncertain situation to solve the uncertain problem also. So as here the battery behavior is a nonlinear one. This method works better and it has given a highest highest reduced error when compared to the other method. So this is the comparison. And finally what we observed was this values we have calculated and this battery SoC value.

Is decreasing the inference we got from this uh uh, data was, if we operate the uh, vehicle, the battery based vehicle at a higher temperature uh, the SoC will decreases fastly. And one more thing is if we, if the driver behavior is aggressive, suppose what a sudden lane change, sudden break, everything means that time also the SoC value is decreasing fastly. And one more thing is the SoC is. There, there is a nonlinearity.

The less SoC values. So if suppose for example if the SoC values less than 20 there the discharge rate is in a nonlinear manner. It's not in a linear manner. These are the observations I have observed in this paper in this work.

Uh, here? Oh, sorry Sir.

Your project on SoC estimation for electric vehicle batteries demonstrates a strong application of machine learning methodologies for a challenging and real-world problem. Now, you mentioned using Gaussian Process Regression due to its probabilistic nature and ability to handle uncertainties. How did you handle the computational cost of Gaussian Process Regression in this context, particularly given its

Take your time, Professor.

Secure in a Gaussian based. Also we had used different kernels Sir. So that kernel value different kernels we have applied over the image. So in that one kernel has given a better accuracy, better value when compared to the other. And here optimization was also done by we have done hyper hyper tuning also so that is used for optimizing value for suppose for K&N. North the number of near number of neighbors is the optimal value if by changing that one we.

Can either increase or decrease the performance of that one. So the same here some kernel value, kernel values five by two kernel was used for Gaussian process and that works better when compared to the other side.

And one more, sorry, one more thing here is any in this Gaussian process regression, all variables, that's all the input variables are considered as an independent one. Then it started its procedure.

Thank you.

Let us move forward. Could you elaborate on your ability to teach both theory and laboratory courses? Specifically, how do you ensure students grasp complex concepts effectively in both settings?

Uh, we can explain the machine learning algorithm from the basic to the students. Suppose we can say everything with an example. For example, if I want to do classification means. What is the difference between classification and regression? First I have to inculcate that knowledge to the students. First. Classification means the output will be in a categorical form. Regression means the output will be in a numerical form or in a continuous form. Then I can start with the basic algorithms. So basic.

From the basic algorithms like linear regression for regression, logistic regression for a classification based 1. So from that basic concepts I can able to deliver my concept and I can make the students to understand also and I can give the best example for that one. For example if our classifications means main. Sorry, mail class with spam or not spam, that classification example I can give or else for many any medical applications I can say whether the patient health is good or bad that I can give from any what any true or false conditions. That analog can give us an example for classification. In case of regression, it's a numerical value. The output is a numerical 1. So for that I need to give some examples like saying I can take a medical application also. So instead of saying the health as healthy, unhealthy.

You can see what is the percentage of healthy healthiness in the patients that, that information I can give. So in the first case, it was, uh, healthy unhealthy, whereas in the regression case, I can say in percentage. So how much healthy he is, how long he will take to recover that information I can give. So by giving a best example from simple to the complex example, and I can make the students to work also to solve some problem also. So in that way, I can teach the students in.

With the simple examples and also I can make the students, I can use more, uh, more think, uh, more questions to solve uh, to the students to solve it. In that way, the students ability also will increase thinking ability also will increase, so they can able to relate the input variables with the output variables.

Thank you.

Could you now provide an example of how you evaluate students and ensure fairness and consistency in your assessment methods?

Sir, can you repeat the question?

Of course, Professor. Could you provide an example of how you evaluate students while ensuring fairness and consistency in your assessment methods?

Uh, here we can evaluate the students in two weeks. Generally, the conventional ways mark. So based on the mark, we can evaluate the students. So instead that mark is not the only criteria, we can give some real world problems to solve. So like tutorial classes, we can give in the tutorial classes, we can make students to to what to show up their ideas and ask them to give some critical problems.

High level problems and we can ask them to think what should be the input and 1st I should not ask the answer first I have to give what should be the input and what should be the probable output. From that we can able to judge whether the students can able to understand the concepts or not. Because if they know what is input, what is output it will be easy for them to understand the concept. So in that way and another one is we can give some.

Project based classes. So instead of uh, what? Instead of conducting the test regularly, we can give one real time problem. It can be a group project. If it is AB tech level, we can give a group project. If it is at M tech level, we can give an individual project and we can make we can give some duration of time as we can ask them to present that also. So this is one way of evaluating and evaluating their knowledge in the subject.

Understood, Professor. Your approach highlights your commitment to evaluating students holistically, not just by marks but through application and critical problem-solving as well.

Can you repeat, Sir?

Certainly, Professor. Your approach demonstrates a commitment to evaluating students holistically—not only through marks but also by testing their abilities to apply concepts and solve critical real-world problems.

Thank you, Sir.

Let’s proceed. Could you describe your experience guiding student research or projects? How do you motivate and direct students throughout such endeavors?

OK, so I can say with an example. So now the students are doing project. Finally students are doing projects. So here first what I will do is I will teach them so they may don't know the concept. Also first I will ask the student I will not give the title. I will ask the students to search the title so it can be in their area. So first they will come up with the subtitle. Then I will say the possibility of completing that project within a stability time so that we will discuss. So here.

In a research point of view we have to discuss a lot so that that from that we can get more idea even as a mentor, even as a students they also can get a more information and after that one the possibility we will check. Then problem identification is very important here. So whatever already some researcher already done but what is our objective? What is our problem identifications that first we have to check. So after finding our project identifications then we can go for data corrections. So after collecting the data, we I first I'll give the schedule to the students. So first they want to identify the problem. After that they need to collect the literature paper. After collecting the literature paper, the main thing here is input. That is the data. Here the data collections where they need to collect the standard data, everything they have to check. Then after that one, we will finalize the data. So with the data we will proceed further and we will get the result within a given time period. The main thing here is the problem identifications. First we have I have to.

With the students to identify the problem. After identifying, we will proceed with the UH from the reference.

Thank you, Professor. Your structured guidance reflects a diligent approach to mentoring student research, ensuring they independently contribute while staying aligned with defined timelines and objectives.

Am I audible, Sir?

OK, Sir.

Could you explain your approach to communicating complex technical topics clearly and structuring your lessons effectively for students?

Yeah, sure, Sir, here.

For explaining this.

Uh, so for, uh, explaining the students, we can explain in a two different way. One is theoretical way, another one is problematically. So first we can explain the students the concept. First we need to explain that clearly. We want to make the students to understand the concept clearly. Then based on some mathematical model, mathematical concept, we can ask, we can solve the problems. Then we can give a few problems and we can ask the students to solve by themselves.

Uh, any prediction based, suppose if I am teaching, uh, prediction prediction, that is regression based model or classification based model means first I will say the concept. After that I'll solve a problem in a board or in a PPT. Then after that I'll give some real world problem. Then I will ask them to think and I'll ask them to solve in the class. Then I'll check that also. So in that way I can improve.

For example, uh.

Thank you, Professor, for sharing your methodical approach to teaching. You're ensuring that students not only understand the theoretical foundation but also gain practical problem-solving skills through hands-on learning.

Sorry to interrupt, we're down to the last 5 minutes.

OK, Sir.

Your resume mentions significant research publications. Could you highlight one publication and discuss its unique contribution to the field?

Yeah, I can say the one publication already I have explained. I will say about another 1. So another one is health monitoring, so health.

OK, OK. Health monitoring, battery health monitoring. So here as like state of charge health monitoring is also very, very important. So that also can be done and here we have used some.

This also can be done.

The internal resistance in batteries, internal resistance, initial resistance, everything was considered for calculating the state of health. So generally that state of health means it indicates the condition of battery when compared to the new battery. So the health condition of current battery when compared to the newly unused battery. So those two will be used for calculating the SOH value. Here some machine learning algorithms can be used by taking this internal resistance, initial resistance.

Using the data-driven based model. So here some parameters in the existing method, uh.

Input and the SOH can be predicted here. So this will ensure the battery safety long utilized span of the battery under the IT also ensure the range and exity will also be reduced here. So this is 1 method. So another one paper. So I'll speak about the image processing paper. So where I have measured the spindle radial error measurement, so generally the spindle.

Radial error measurement can be done by using some capacitance based or inductance based method.

In mind I have used image processing. I have included image processing here for measuring the spindle error message errors. Because in the sensor based method the main problem is the sensor has to be placed near the spindle at very small distance but that may at the highest speed that may affect the sensors also the capacity or inductive sensors also. But in my method I have used a camera images was captured for different spindle speeds the different.

The uh, the sequential images was captured and from the sequential images, the spindle, umm, radial distance, radial error was measured. Generally, uh, the center point, the spindle center point. Is measured from the center point. The radial error was measured and there are generally 3 errors accumulated over this radial error. 1 is centering error, synchronous error, and asynchronous error. Centering error means there is a centricity under synchronous error. Error means that's a periodic error that will.

Fix the surface whereas asynchronous error so that will affects the final product obtained from the spindle. So it has to be taken care in our observation if this spindle speed increases the synchronous error also error got decreases that is observation we have been there.

Thank you, Professor. Both works—the battery health monitoring using machine learning and the spindle radial error measurement through image processing—showcase not only technical depth but also significant innovation in addressing practical challenges.

Thank you, Sir.

You're welcome, Professor. Could you briefly describe your experience in collaborating with industry on projects or consultancy work?

As of now I'm interested to collab to collaborate with industry and do consultancy project also but as of now.

I have received a fund for conducting UH-2 week UH faculty development program UH from ACT for an amount of UH 5,80,000 and I have successfully completed in the year 2020, academic year 2021.

Uh, also I have submitted one patent.

In the year 2122 I have submitted projects but I haven't received any. Any received any project last year. Also I have submitted 1 project to ICSS that is related to.

Battery health.

Thank.

Finally, could you elaborate on how your qualifications—including your PhD and certifications—specifically enhance your ability to contribute effectively as a professor in Artificial Intelligence and Machine Learning?

Shows as I have conducted one ACT FDP program on.

Machine Learning and Deep learning In the year 2021 I got interest in doing project in machine learning. Machine learning based projects. So if it is a So first I started my work by collecting the data from the Google repository Scaggle UCA. Then I started proceeding with that one. There I have learned how to work with machine learning and deep learning algorithms to improve my knowledge. I have also done recently I have done a six months hybrid mode.

In the Indian Institute of Information and Technology, Trichy in the title, machine learning and deep learning, sorry, machine learning and cyber physical systems and I secure 9.3. So this certification will help me to handle artificial intelligence and machine learning students as well Sir.

QAP program that is also sponsored by a act only from.

Thank you.

Professor, to begin, could you share which foundational topic in Artificial Intelligence or Machine Learning you particularly enjoy teaching, and why?

1st I started with artificial intelligence there it's a it's a theory based 11 only the theory part was there. Then after that I have taken machine learning concept also. So in a machine learning it's an algorithm based 1. So there I can give more example for students also. So I was. Very eager and I excited to take the machine learning class where I can give some examples for them to to make them understand and that will.

Help myself also to learn more and I it will help help the students also to to learn a lot Sir.

I see, Professor. It’s excellent that you value creating practical connections for your students through examples in machine learning. Providing relatable examples certainly enhances understanding. Thank you for sharing. Let's proceed.

Professor Kavitha, I can see your background includes extensive research in areas like Machine Learning and Deep Learning, particularly as reflected in your role as a Principal Investigator for a funded project titled "Machine Learning and Deep Learning - An Application Perspective." This aligns closely with the kind of expertise we value highly. Can you explain the key objectives and findings of this project? Specifically, how do you think they translate into practical applications for individuals learning about these subjects?

I see so this FDP was conducted in the year 2000, 2020 one there only most of the people started working on machine learning and deep learning and there what they observed was it's a title is machine learning and deep learning and application perspective. So there we got researchers from many places, many standard universities they're what I observed was the machine learning concept can be used for different variety of applications. So at the basic.

The my knowledge was this artificial intelligence machine learning will be only for CC people. But it was not like that. So this machine learning and deep learning concept can be used by electrical people, mechanical people, electronics people, computer science people, even biological people. They also can use this machine learning. So that observation I understood. Then I got eager to learn this machine learning and deep learning algorithms. So it can be used. It's a widespread 1. So it can be used by any people.

So it can be used by chemistry people also any uh, science and humanities people also they can use if they have a proper data, good data, they can go for this machine learning and deep learning, this concept I understood from that one then only I started doing my project in this.

It’s intriguing that your project highlighted such interdisciplinary applicability of machine learning and deep learning, going beyond traditional computer science fields to areas like biology, chemistry, and even humanities. This insight, as you mentioned, opens doors for researchers from various disciplines to leverage these advanced methodologies if they have structured, meaningful data available. How do you think this interdisciplinary discovery influences your ability to effectively instruct students from diverse academic backgrounds about these concepts? For example, how would you adapt your teaching methods to ensure students from non-computational sciences grasp the principles of machine learning effectively?

Time is up! Ending the interview now.