Anjith George is a research associate in the biometric security and privacy group at Idiap Research Institute, focusing on developing face recognition and presentation attack detection algorithms.

His research interests are real-time signal and image processing, computer vision, and machine learning, with a special focus on Biometrics.

INDIAai interviewed Anjith to get his perspective on AI.

How did a graduate in electrical and electronics engineering begin AI research?

My graduate degree was in Electrical and electronics engineering. Later, I completed my master's degree in the electrical engineering department at IIT Kharagpur, with a focus on instrumentation and signal processing. My master's thesis was on computer vision, and I learned the fundamentals of computer vision and machine learning during this time. Back then, the entire field of pattern recognition was undergoing a significant transformation. Interaction with lab peers gave me an early taste of research. I became fascinated by machine learning and pursued a PhD in computer vision.

Describe your doctoral research area. What are your significant contributions to research?

My doctoral research focused on developing image-based eye-tracking algorithms and applications that use eye-tracking data. Eye gaze provides a direct indication of a user's attention. Nonintrusive eye gaze tracking has numerous applications, including fatigue detection, biometric authentication, alertness and activity recognition, human-computer interaction, and so on. Unfortunately, existing eye trackers were prohibitively expensive, limiting their practical application. To that end, we created various gaze-tracking algorithms using the low-cost webcams and head-mounted cameras.

Furthermore, we proposed a biometric identification method based on eye movement data to demonstrate the benefit of using eye-tracking data. The proposed method achieved state-of-the-art performance and could be integrated into existing iris recognition systems to make them more resistant to spoofing attacks. In addition, we proposed a framework for activity recognition that uses gaze data from a head-mounted eye tracker. We demonstrated that combining data from gaze, ego-motion, and visual features improves classification accuracy in indoor environments where traditional activity detection modalities fail. This research could be beneficial in both virtual and augmented reality applications.

How do the research practices of Indian universities differ from those of their Western counterparts?

The research environments in India and the West are vastly different. These are some observations I've made.

• I have noticed a general disconnect between academia and industry in India compared to Western countries. Most research in Western countries is either applied or still oriented toward industry. It facilitates funding because the issues addressed are relevant and add value. There are also opportunities for students to do industrial internships to become acquainted with real-world challenges. 
• In terms of research outputs, western research labs have made significant efforts to make research outputs open source to lower the barrier to entry and increase visibility.
• In Western countries, tasks and deliverables are typically fixed and frequently aligned with field advancements. In India, the objectives and timelines for PhD students are often flexible, which, combined with low stipends, makes pursuing a research career a tough call. 
• Another critical difference is the availability of resources; in many cases, in western countries there are sufficient computational facilities to meet the needs of researchers, which would be more difficult in Indian universities. 
• In India, there is a lot of red tape that is not conducive to creative environments such as research institutes. 

What were your earliest obstacles in AI research? How did you overcome them?

When I began my PhD, machine learning was primarily concerned with feature engineering and traditional ML models such as SVMs and Logistic Regression. These models were more mathematically grounded, and neural networks were generally dismissed Only a few toolboxes were available to try out or reproduce works from other groups. In addition, There was also a constraint in terms of computing resource availability. Deep learning was more of a black art that was only accessible to a select few. I recall running experiments on my laptop with an old graphics card. Many companies and labs are now open-sourcing their deep learning frameworks, lowering the barrier to entry.

What tactics do you employ to maintain your composure when an experiment fails or an article is rejected?

Perseverance is an essential quality for researchers. Most new ideas do not succeed the first time they are tried. It is usual for experiments to fail, but each failure teaches you something. Investigating why things fail yields insights that were not obvious at first. 

What is the one thing that every researcher must consider?

Develop a genuine curiosity towards what you do. After all, "research is just formalized curiosity". Having this perspective encourages creative thinking and makes work more enjoyable. Always be on the lookout for new ideas and keep an eye on how the field evolves.

What advice do you have for aspiring to work in artificial intelligence research? What should they concentrate on to advance?

I advise anyone who wants to work on AI research to hone their programming skills and gain a solid understanding of the fundamentals of linear algebra, probability, and statistics. With this, you are in a good shape to start exploring the current AI landscape. The field of AI is going through a phase of explosive growth, every other day you find a new AI breakthrough. It's exciting, but it's also becoming more difficult to keep track of. Start reading literature about the basics to develop your intuition. Reading research papers, following blogs, and participating in forum discussions are good ways to stay current. Also, machine learning should be hands-on, implement existing works, and try out example problems. Competition websites like Kaggle provide an excellent opportunity to develop skills.

What scholarly articles and books have significantly impacted your life?

For machine learning, Pattern Recognition and Machine Learning by Christopher M. Bishop was the default reference. "Deep Learning" by Goodfellow is another excellent book that covers a lot of recent approaches. Aside from machine learning, one of my childhood favourites was Yakov Perelman's "Physics for Entertainment".