While we all agree that the field of Artificial Intelligence holds immense potential, it is also susceptible to malicious attacks which could severely limit its capabilities in mission critical domains such as healthcare !

Vishal Rajput, our next pathbreaker, Senior AI Engineer at SkyeBase (Belgium), works on state-of-the-art AI models and techniques to create a comprehensive AI pipeline used for inspecting cranes at Sea Ports.

Vishal talks to Shyam Krishnamurthy from The Interview Portal about his foundational research on Adversarial Attacks, focused on increasing trust in the AI systems for medical data.

For students, don’t just have the motivation to learn AI, instead apply it to address bigger challenges that various industries are facing.

Vishal, Your background?

To begin with, I’d like to introduce myself briefly. I hail from Noida, where I spent my childhood and adolescence. While I excelled in academics and the arts, I struggled with sports. My father worked in accounts, and my mother took care of our family and home. During my school years, I followed a similar path as my peers. However, it wasn’t until I took a gap year to prepare for the IIT that I realized the true challenges that awaited me. Though my initial passion was singing, I now use it as a way to unwind after work.

What did you do for graduation/post graduation?

Due to the limited options provided by JEE, my B.Tech in Electronics and Communication at IIIT Jabalpur was solely based on my rank. However, after just one semester, I realized that the course material wasn’t going to take me very far. In response, I began working on numerous Embedded Electronics projects independently during my first year. By the end of my second year, I had dived deep into python and started to develop an interest in Image Processing and IOT. However, due to financial limitations, I was unable to purchase new hardware for each project, leading me to shift my focus towards software and eventually to AI. At the time (2016), AI was a relatively new field that many people were unfamiliar with. While it may seem like a stroke of luck, there was certainly more to it. In my pursuit of finding my passion, I explored many other areas, such as management, digital marketing, front-end development, competitive coding, and more, before finally discovering my love for AI. And later on, I decided to pursue a Master’s degree in AI to formalize my knowledge of the subject. In 2022, I completed my Advanced Master’s in AI at KU Leuven. Few of you might think, but why Belgium for masters? I did have a lot of other universities to go to on my list, but Belgium offered the lowest tuition fees and a world class education (KU Leuven ranks under top 50 in the world) which I was able to completely afford on my own without any scholarship or loan.

Tell us, how did you end up in such an offbeat, unconventional and cool career?

It all began with a simple YouTube video on AI that Google suggested to me. The video featured a basic classifier, and I was so impressed by it that I attempted to write the same code seven times. At first, I didn’t understand the code, but with each try, I gained a better understanding of what each line did. This was my official start of my journey in AI. From that point onward, I delved deeper into AI-related topics. Initially, Harrison Kinsley from Sentdex helped me establish a strong foundation in Python and the basics of AI. Later on, I discovered other YouTube channels, such as Deeplizard for coding AI, 3blue1brown for deep mathematical understanding, and Statquest for statistics, among others.

During my first two years of exploration, I wasted a lot of time sifting through poor quality content to find the good stuff. I learned that investing time in filtering out poor content is necessary to acquire a good understanding of any topic. By my third year, I was certain that I wanted to pursue a career in AI, and so I heavily invested myself in computer vision and image processing research. I wrote seven research papers in AI, data analytics, and image processing for various journals, including SCI journals. Though I struggled with electronics subjects, I excelled in AI research.

I then joined a course from Applied.AI, which helped me put my theoretical knowledge into practice. Ultimately, I decided to pursue a Master’s degree in AI, and my path was nearly set. However, there was another component that shaped my journey: my interest in human behaviour, psychology, and philosophy during my bachelor’s degree. Studying these subjects helped me broaden my perspective and approach to new ways of learning. Without studying philosophy, I would never have ventured into the uncharted territory of AI. Even now, I consider myself a thinker and philosopher before an AI engineer.

How did you plan the steps to get into the career you wanted? Or how did you make a transition to a new career? Tell us about your career path

Before starting my professional journey, I worked at Delhi Technological University (DTU) as a researcher. I explored the topic of OCR or Optical character recognition. After months of hard work and thinking, I was able to come up with a novel algorithm. In those 6 months, I wrote 3 research papers including 2 papers on underwater Image enhancement and Image watermarking. For my 6 months at DTU, I went into a hibernation kind of situation, I worked every single day from 9AM to 9PM without any breaks, I barely talked to anyone including my parents or friends. Before I finished my bachelors, I produced 7 research papers, with the help of Dr. Irshad Ahmad Ansari at IIIT, Jabalpur and Dr. N. Jayanthi at DTU. I’m glad that I pushed myself hard to know my limits and later on this helped me in finding a good balance between work and life.

Next thing was a brief stint at Capgemini, I didn’t learn much except for a valuable lesson about the inner workings of companies. I realized that I needed to learn more to make a complete switch to AI, so I decided to pursue a Master’s degree in AI. Meanwhile, I continued to work with my university professors on AI research. 

While at KU Leuven, I joined the UZ Leuven medical Imaging Centre and conducted foundational research on Adversarial attacks, which led to the publication of a paper. This experience taught me a great deal about AI/ML platforms such as Scikit and TensorFlow. Let me summarize the research I did at UZ Leuven. All types of Neural networks are very easy to fool, all you need to do is add some calculated noise (noise is added through calculating gradients) to your input and voila, your network will identify a car as bird and bird as some other random thing. My work was focused on solving this issue, increasing trust in the AI systems for medical data. If a machine says its cancer looking at some scan, it should also tell why it thinks that it’s cancer. We all know that medical data is critical and doesn’t have scope for errors, thus my work made all types of neural networks more safer and explainable.

After completing my Master’s degree, I worked as a research trainee at SONY R&D, where I gained insight into the financial side of AI and the difference between academic and industrial research. My work at SONY focused on identifying credit fraud and loan applications; building a framework where we trained models to handle adversarial attacks. Bank use a lot of features to decide whether to give you loan or whether this transaction is real or not. Out of 100s of features these models uses, something like number of pets or your pin code can change your loan request. People use these vulnerabilities to scam banks and we created a framework to solve these  vulnerabilities and save money for the bank.

Later, I joined SkyeBase as an AI-Vision Engineer, where I am responsible for all AI-related development, from strategy formulation to deep AI research. Additionally, I started my own blogging site, AIGuys, where I have written 63 AI-related articles and a total of more than 125+ top-rated AI articles. My goal is to bring the most complex ideas to the community in plain and simple English.

My approach to learning is simple: invest time and be patient. There are no shortcuts, no single mentor or course that can teach everything. It’s a combination of everything that helps live and breathe AI every day. My social media feeds are only filled with AI-related content. On an average, I watch 8-10 hours of AI, psychology, philosophy, and neuroscience lectures every week. I believe in learning how to learn new things and doing it on my own. I also prepare for the possibility of team members backing out or underperforming in AI projects. To feed my mind, I avoid consuming things that are just a waste of time. Overall, my approach helps me build a comprehensive picture of what I’m doing and why I’m doing it.

How did you get your first break?

I knew perfectly well what I was doing and why I was doing it. I didn’t have any break, it’s just sheer perseverance that got me my current job. 

I had to face 240 rejections before finally landing my current job. Most of the rejections were due to my lack of industry experience and due to visa requirements, rather than my AI knowledge.

What were some of the challenges you faced? How did you address them?

Challenge 1: My only challenge in securing a job in Belgium was my lack of European citizenship. It was incredibly difficult to find an employer who was willing to sponsor my visa, regardless of my skills. I received 240 rejections before finally landing my current job. Every day, I would wake up to 5-6 rejection emails, and there was nothing I could do except continue applying. Most of the rejections were due to my lack of industry experience and due to visa requirements, rather than my AI knowledge. However, I remained optimistic throughout the entire process and told myself that it would make for a great story in retrospect.

Challenge 2: This is not a real challenge, but more of a process, it’s not knowing what’s important. Go ahead and just waste time figuring out things on your own. Everyone has a different temperament, thus it’s better to find your own calling than walking on someone else’s path.

Where do you work now? Tell us about your current role

Currently, I work as a Senior AI Engineer at SkyeBase. SkyeBase is an asset inspection company that owns a suite of drones for underwater asset inspection, indoor inspection, and of course industrial drones for assessing oil and gas pipelines, big cranes and many other critical infrastructure. At SkyeBase, I am responsible for everything related to AI, including combining state-of-the-art AI models and techniques to create a comprehensive AI pipeline used for inspecting cranes at Sea Ports. I have personally collected data on a bicycle multiple times, and plan out the development of the entire module. I collaborate and guide researchers and universities on what we need from them, and I handle all research and implementation tasks while playing a significant role in product envisioning. I constantly explore new topics and spend the majority of my time learning about the latest developments in the field of AI. For any new AI module, I typically spend around 80% of my time researching and 20% actually developing it. I love everything about my work, as it gives me the freedom to build, strategize, research, and innovate.

How does your work benefit society? 

I’m making the process of inspecting ports safer while saving a ton of money and resources that are generally required for inspecting a sea port crane and other critical infrastructure.

Tell us an example of a specific memorable work you did that is very close to you!

Winning the Kaggle competition in my masters was memorable. We outperformed the SOTA AI modes in facial recognition using simpler DL models combined with SVM. 

Your advice to students based on your experience?

Stop thinking about motivation to learn AI because it only stays for a few weeks at best, rather, build a discipline of engaging in AI content. Stop watching crap from social media, use it only for AI and other learning purposes. If you want to know what’s happening in your friend’s life, go ahead and call them instead of following their daily social media stories. If you can sort out  your personal life, it won’t take much to sort out your career and professional life. Once you are deep into AI circles, you will automatically know what the new AI research is and what you need to do to be good at it. 

Future Plans?

Take AIGuys to a bigger audience, do more AI research, guide more students and companies in best AI practices. And eventually, I set up my own AI company.