Dr Apurba Das works as a Head of Cognitive AI at Tata Consultancy Services.

He has published five books and approximately forty articles.

Apurba has been granted 15 patents in the United States, Europe, and Japan out of 40 patent applications.

INDIAai interviewed Apurba to get his perspective on AI.

Could you tell us about your professional and academic background?

I am heading the Cognitive AI group of TCS IoT as the Chief Architect. Before joining TCS, I worked as Technical Architect for HCL and a Scientist for CDAC, Kolkata. I have been awarded for global research collaboration and significant contribution to patents in California, USA, by HCL and Xerox. Recently I have been honoured with the Distinguished Engineer award in TCS.

I have done my B.Tech in Electronics and Communication Engineering, M.Tech. in the specialization of Intelligent Automation & Robotics. Recently, I completed my PhD in Computer Science & Engineering.

How did you get started in the field of artificial intelligence?

My complete experience of 19.5 years is in Computer Vision/ AI/ Machine Learning. During my B.Tech. Curriculum, Digital Signal Processing and Communication Engineering were my favourite subjects. Even before that, I was fond of geometry and even painting. Most probably, the signals and patterns attracted me from the early age of my exploration of the world of knowledge. During my M.Tech, I was formally introduced to the fabulous world of Artificial Intelligence, Computational Intelligence, and Computer Vision. I started visualizing images as a perfect combination of geometry, pattern, and art. When I started my job in CDAC as a Scientist in the Advanced Image Processing lab, my love for signals gave me an added advantage. I have always been passionate about my work, therefore. I always feel if artificial intelligence is providing intelligence to a machine, the final objective of that machine should be to sense and react on its environment intelligently. A device can have multiple sensors like we as humans possess five sensors (vision, smell, hearing, touch, and taste). However, vision is undoubtedly the most important and accurate sensor. I love to design my algorithms and see them working intelligently in solving real-life problems in various sectors. I feel utmost satisfaction when I can enable a dumb machine to learn from mistakes, do reasoning, plan, sort, and recognize objects or subjects of interest. I feel amazed by seeing and becoming part of the AI journey where the machine can improve plant productivity, catch shoplifters, ensure safety, analyze quality, perform complex robotic surgeries, help differently-abled people and many more.

What was the topic of your doctoral research? Can you tell us about the research problems on which you focused?

The title of my PhD thesis is "Fast Range-Domain Filter for Efficient Dehazing of Videos using Deep Learning". There are six classical problems in computer vision:

1. Environmental/ Illumination diversity
2. Availability of data
3. Choice of the exemplary architecture
4. Real-time performance requirements
5. Very high accuracy expectation
6. Privacy

In my doctoral research, I have addressed the challenge of an adverse environment and its impact on video analytics with minimal data dependency ensuring accuracy-performance trade-off optimization. In the first part of the thesis, I designed a fast and efficient bilateral filter which can improve image quality by preserving edges. The contribution showed improvement in speed of computation and quality in image reconstruction. The filter has been designed using a novel deep neuro-genetic network. It has demonstrated excellent quality improvement without any negative impact on performance.

I used the said filter to dehaze videos due to fog and rain in the second part. Also, I have shown how my designed filter can effectively improve the quality of endoscopy and laparoscopy images/ videos.

Who or what inspired you to pursue AI research as an electronics and communication engineering graduate?

I have been inspired by many teachers, colleagues, and friends. To name a few, 

• Prof. Amit Konar from Jadavpur University, 
• Mr Debasis Mazumdar from CDAC, 
• Dr Stephen Dashiele from Xerox. 

To pursue research in AI and computer vision, what maximum inspired me is 

• the beautiful world of patterns, 
• extracting real wisdom from the environment around us, 
• engineering the same in such a way that even a dumb machine can also mimic us, and 
• solve real-life problems in various fields of medical, retail, manufacturing, transportation, etc.

What difficulties did you initially encounter in the industry? Could you describe how you overcame them?

Let us first talk about the technical challenge. Proper data availability and performance-accuracy trade-off are the biggest challenges while designing an AI engine. For AI system deployment to different customer premises, asking for and acquiring other data with all possible variations is another challenge. I have attempted to overcome them by developing innovative algorithms which are not only data dependent. That's why most of my designed systems hybridized deep learning models (data dependent) and traditional image/ video processing algorithms (not reliant on data). Furthermore, I attempted to ensure that any new deployment would adapt to the new environment automatically without any additional data feed.

There are other difficulties in AI adaptation by different industries, especially entry-level workers, as they feel the AI would do their work and be fired. I tried to convince the stakeholders that AI would assist humans in doing their job faster. Even if AI can completely take care of complete execution, we would create new and better kinds of jobs for them.

What are the most widespread myths you would like to dispel as a long-time member of the AI and machine learning community?

The two most widespread myths about AI are: 

(a) AI will take over our jobs, 

(b) AI will control the world.

I do not think AI can take over our jobs. The same myths started spreading when the personal computer was invented. AI would instead help us to do our job more efficiently. So most probably, we will start doing our job in a smarter way differently.

AI can't control the world because that kind of national security-related decision would never experiment with machines. At most, machines can make recommendations and suggestions based on available data points. The decision would still be the privilege of humans.

Tell us about your responsibilities as Tata Consultancy Services' Head of Cognitive AI. What is the routine of your workday?

As a leader in information technology and the Internet of Things (IoT), Tata Consultancy services serve customers through contextual knowledge and utmost rigour. My team is responsible for developing AI solutions to strengthen TCS's intellectual property and serve customers to solve real-life problems. My team designs, create and delivers AI solutions to customers across different industry verticals like retail, healthcare, manufacturing, transportation and logistics. My job is to interact with other customers of TCS to understand their pain points, define their business problems as legitimate technical problems and design the solution for the same. I am responsible for running a large team of 54 members and improving the group's AI business.

It's great to know that you've been granted 16 patents. Could you tell us more about those patents?

To date, I have filed 41 patents in various geographies, of which 16 have been granted. All my patents are related to AI vision, applicable to different domains ranging from document image processing to colour science, autonomous vehicles, 3D reconstruction, adaptive binarization, surveillance, multi-camera tracking, face recognition, aerial image analytics etc.

What advice do you have for those who want to work in AI research? What are the most efficient methods of progress?

My advice would be to start seeing the world as it is. If you start enjoying your life journey with utmost involvement in everything, you would love to create a software agent who can perceive and react intelligently to the environment. Identification of potential problems is the most critical task. Identify a problem which can help common human beings and start exploring the possibilities to solve the same using AI.

Could you provide a list of notable academic books and journals on artificial intelligence?

1. IEEE Transactions on Artificial Intelligence

2. Amit Konar, "Artificial Intelligent and soft computing", CRC press

3. Apurba Das, "Guide to Signal and Pattern in Image Processing", Springer

4. Artificial Intelligence, Elsevier

5. Stuart Russell and Peter Norvig, "Artificial Intelligence: A Modern Approach."