Hey everyone! I am Kunind Sahu, a BTech sophomore in the MEMS department (to-be-thirdie :/ – I was forced to leave campus in my first year due to COVID and now I’m a to-be-thirdie. I am in denial). I am pursuing a minor in Artificial Intelligence from CMInDS, IIT Bombay and also heading the Data Analytics and Visualization this year! You would often see me binging on Horror and Thriller shows/ movies and also playing video games (really loved the Uncharted Series) in my free time.
I’m currently interning at Delft University of Technology, Netherlands (known to many as – TU Delft) in the field of Artificial Intelligence.
My motivation to pursue Artificial Intelligence
It was during my summer vacations, after my first year had abruptly ended due to COVID when I first delved into Artificial Intelligence. I had a lot of time to kill and there was a lot of buzz about AI as a field – which intrigued me quite a lot. So I decided to go ahead and give it a try and I did the famous Machine Learning Course on Coursera offered by Andrew Ng. I really liked the course structure and found the field to be genuinely very interesting. This led me to make up my mind about pursuing an AI Minor and learning more about the field in general.
My motivation to pursue a Research Internship
Ans) I had done a project in Data Science for the course DS203 and I thoroughly enjoyed doing it. This, coupled with the fact that I would not want to waste my summers and also the fact that I was unsure as to what I wanted to do later on (an MS/ Job), really made me gravitate towards a Research Internship. I would be working on a novel problem and having an early exposure to research would really help me with my MS applications if I later would want to go forward with it.
Also, I had a chat with a few seniors of mine (did some research on my own too) – about the applications of AI Materials Science. To my surprise, I got to know that AI was catching steam in the Materials Science field as well – in various applications such as Novel Molecule Discovery, Drug Discovery, Predicting results of simulations etc which got me all the more excited and made it clear to me that a Research Internship was the way forward for me.
My Apping Experience
I had made up my mind midway through my winter vacations in my second year that I would like to go for a research internship. But I knew for a fact that I was just a second year undergraduate student with no formal research experience, hence doing a research internship in Artificial Intelligence with a CS Professor would be really tough.
Hence, I decided to look for Materials Science/ Mechanical Engineering professors who were working in AI and began databasing (i,e creating a list of professors – along with their research interests, their websites and their email IDs). This took me some amount of time because there were not many professors who had explicitly stated Machine Learning as one of their research interests.
I mostly went through the Materials Science faculty list of various colleges (based on their ranking). The entire databasing process was very exhausting – I could only find about 40 professors whose research interests matched mine.
Now was the time to send them mails requesting for a research internship. When applying for a research internship via mail, it is essential to keep your mail brief and to the point. Also, try to use tools such as Streak to track your emails. Moreover, do keep in mind the time difference between countries while sending out the mails.
I had applied to about 20-25 professors out of which only 5 replied, and out of those 5 only 2 were interested.
Specifics about my Internship
All the AI enthusiasts out there would be delighted to know that TU Delft has about 24 Labs exclusively dedicated to AI and its applications. I am working in the MACHINA Lab headed by Prof. Miguel Bessa. I’m working on the problem of Graph Similarity Computation. Graphs are very general data structures and calculating the similarity of two graphs is an NP-Complete Problem. No known algorithm can reliably compute the similarity between two graphs for graphs having 16 or more nodes. That is why we turn to Deep Learning – or more specifically Geometric Deep Learning where we apply Deep Learning Techniques to unstructured data such as Graphs and Point Clouds. Graph Similarity Computation has a lot of applications in Neuroscience, Computer Hardware Security, Computer Vision and Computational Chemistry & Biology to identify similar Chemical Compounds. We are trying to improve upon current Deep Learning models to compute Graph Similarity.
To my dismay, TU Delft was closed on account of COVID, hence the mode of the internship was remote. When I had applied for my research internship, I did not have any particular topic on my mind, I just wanted to enjoy and learn new things related to AI and Applications of Deep Learning. When Prof. Bessa first proposed the project idea, I got a bit nervous because Graph Theory and Geometric Deep Learning was something I had never ever heard of before, but after giving it some thought, I decided to go ahead with the project topic because it seemed very interesting.
I was working in tandem with the professor and another PhD student of his who was from Russia. He was very approachable and gave me a lot of autonomy about deciding how I wanted to move forward with whatever ideas I had. He wanted me to first review the current State of the Art Algorithm for Graph Similarity Computation, try to critique each of its architectural choices and understand the impact of each pipeline in the algorithm, and then actually try to improve upon it by making certain changes in its architecture.
We have biweekly meetings where we discuss whatever work I have done and try to plan out what we would do in the following weeks.
The work from home setting was certainly a bit disappointing because I missed out on being able to visit TU Delft and have an opportunity to meet with the group face-to-face, but the online experience was not bad either. I really would have loved to visit Europe – I mean who would not have XD.
Me during the online Spring Sem / Summer without the internship
Me during my internship (can’t deal with another online setting ugh :/)
Challenges I Faced
The learning curve was steep and the project was a bit challenging because I had never worked on the problem before. I was completely new to it and had to brush up on a lot of things – which included learning about Graph Theory, learning an entirely new library for creating and manipulating graphs and also to implement Deep Learning Algorithms on graphs.
Also, I’d have to point out as well that the data which we required for our task was not available in a viable format so, I had to perform a lot of feature engineering just to convert the data into a viable format (since we fed a pair of graphs into the model). Another challenge I faced was parallelizing the algorithm – since I was working with a pair of graphs, it was not as straightforward to actually create batches and implement parallelism to help the model perform the computations quickly. But after hours of reading the documentation of the libraries, I certainly was able to implement it!
My research internship is still going on – but I have to say that it really helped me undertake independent research and helped me exercise my mind as to what changes I could make in the current model. It exposed me to new tools and technologies and a completely up-and-coming area of research. I really liked the field and would certainly like to learn more about it! I would even consider pursuing it for my MS as well – if I finally decide to go forward with it.
What I would like to say to you all is – try to have fun and learn as much as you can during your research internship. Network with the other people in the group as much as you can. I can safely say that applying for a research internship was one of the best decisions I have made and I have no regrets whatsoever. I’d really like to thank Insight for giving me this opportunity to talk about my research internship. Feel free to reach out to me in case of anything!