Ramanarayan Mohanty works as a research scientist at PCL India-Intel Labs, where he develops graph-based machine learning and deep learning algorithms.
His research interests are mainly in graphical AI, machine learning, deep learning and computer vision.
INDIAai interviewed Ramanarayan to get his perspective on AI.
How did your interest in algorithms come about?
While pursuing my MSc at IIT Kharagpur, I met some of my seniors pursuing their PhDs and working on machine learning algorithms. During our tea time, we had a few discussions, which motivated me to work on algorithms for my PhD.
Could you tell us about your professional and academic background?
I graduated in Computer Science and Engineering from Biju Pattnaik University of Technology. Then I pursued my MSc (MS Research) in Embedded Systems and PhD in Graph Machine Learning at Indian Institute of Technology Kharagpur. After completing my PhD, I joined a company called Pathpartner Technology as a senior research engineer. Then I moved to Intel Labs, where I work as a research scientist.
What differences and similarities do you notice between being an academic researcher and a scientific researcher?
In India, the significant differences between an academic researcher and a researcher in industry are as follows:
i) The industrial researcher has a very well-defined problem from day one, but for an academic researcher, the problem statement is not as clear and well-defined; over time it becomes clearer.
ii) As researchers we have an abundant number of resources in terms of machines, clusters and whatever we need, we get them immediately, but as academic researchers we don’t have that luxury.
iii) Another important difference is that in industry most of the research results are directly applied in real life scenarios or real applications. Yet in academic research, this is not always the case.
Likewise, we mainly opt for reviews and conferences in industry and academia. Both in industry and in academia, we carry out both theoretical and application-oriented research.
Tell us about the problems you solved during your doctoral research and the topic you chose.
During my doctoral research, I worked on hyperspectral image classification problems in adverse and harsh environments. A problem I have worked on is the “class identification and discrimination problem” in hyperspectral image classification. Hyperspectral images are captured by satellites or drones containing hundreds of spectral bands. Each image includes a few square kilometers of space on earth, including houses, roads, forests, bodies of water, green fields, etc.
Each entity is called a class, each with a unique spectral signature. Spectral signatures are the main discriminating feature of hyperspectral images. However, sometimes two classes have the same signature, such as the road and the roof of a building have the same spectral signature because they are made of cement. This creates confusion in image classification, known as the “class discrimination” problem.
Similarly, a class of images sometimes has two different spectral signatures due to the variation in light. This problem is called the “class identification” problem. To solve these problems, I used techniques based on machine learning of graphs by considering both spectral and spatial characteristics. Likewise, I had worked on other interesting problems in this area.
Tell us about your role as a Research Scientist at Intel India Parallel Computing Labs. What is your daily routine?
At Intel, the work culture is very flexible. As a research scientist at Intel Labs, my daily routine involves the following:
- Read research articles or blogs.
- Performing experiments and analyzing results.
- Attend some meetings and do some project planning.
In addition, I spend time at foosball and in the cafeteria to relax my mind. All in all, the day is quite pleasant and, at the same time, productive.
The global computer vision market was worth USD 11.22 billion in 2021 and is projected to grow at a compound annual growth rate (CAGR) of 7.0% between 2022 and 2030. Additionally, artificial intelligence (AI) in Computer vision technology is gaining popularity in various applications. What do you think of this market trend?
Computer vision (CV) with deep learning (DL) is one of the most progressive and fastest growing areas of research and application. Advances in resumes with DL search immediately adapt to the commercial world. In current scenarios, CV and DL are massively transforming industries such as security, retail, robotics, manufacturing, automotive, healthcare, and more. As you might expect, this commercial use of resumes is just the tip of the iceberg. The full potential use of the CV is yet to come. The rapid development of augmented and VR apps like Metaverse is making CV more intuitive. It integrates into people’s lives by allowing them to interact with real objects in the virtual world. The future of CV and DL will be significantly impacted by new architectures, advanced cloud solutions for managing big data, and automated solutions to further reduce time to market. The recent advancement of transformer architecture in CV and cloud computing services will provide further impetus for scaling population-scale CV and DL solutions with enhanced capabilities. It will provide an upward push to this trend.
What advice do you have for those who want to work in AI research? What are the most effective progression methods?
Currently, AI is a hot topic; everyone wants to pursue a career in AI and data science. Therefore, the competition is also very high. So, to pursue a career in AI, students need to look into two fundamental things: proper programming skills and a fundamental understanding of machine learning concepts. From a programming skills perspective, they should be proficient in any programming language (for example, Python is most often preferred). From the perspective of the ML concept, they must know four pillars (linear algebra, probability, statistics and calculus).
I think learning AI by doing is the most effective method of progression. Either reading research papers, trying to understand them and implementing them to replicate the results, or implementing small capstone projects is the most effective way to progress.
Could you provide a list of notable academic books and journals on artificial intelligence?
There are many AI books and journals available in the market. However, I found some very interesting books,
Bishop’s “Pattern Recognition and Machine Learning” are very good for fundamental machine learning,
“Dive into Deep Learning” by Alex J. Smola is an excellent open source interactive deep learning book available for free on the internet.
Other than that, Goodfellow’s “Deep Learning” is also a great book to follow. From a research perspective, many journal and conference proceedings are available. Some of the most advanced and notable conference proceedings include Neurips, ICLR, ICML, AAAI, CVPR, ICCV, ECCV, etc. The published articles are free and anyone with an internet connection can easily access them.