When you are thrown in the deep end in the beginning of your career, you develop resilience for life that helps you thrive in dynamic and entrepreneurial environments.

Arvind Srinivasa, our next pathbreaker, Senior Research Officer at Singapore Bioimaging Consortium (SBIC) A*STAR, Singapore, applies Image Recognition and Deep Learning techniques in Biomedical Imaging for effective prognosis in Healthcare.

Arvind talks to Shyam Krishnamurthy from The Interview Portal about his first exposure to machine learning in addressing a real problem of flood assessment and his initial struggles in developing ML algorithms based on satellite images that sealed his curiosity and fascination for applying Deep Learning to solve real world challenges.

For students, the opportunities to learn always come in the early stages of your career, when you have the drive and the will to master the concepts. Always take up what comes your way as well those that don’t come your way !

Arvind, tell us about your background?

I grew up in a small town called Channarayapatna (Karnataka) and later in Bengaluru,  which in 90’s was called “Pensioners Paradise”, and now “Silicon Valley” of India. My father worked for New Government Electrical Factory (NGEF), and my mother was a primary school teacher. As a child, I was brought up in an orthodox family with values such as honesty, integrity, giving respect to elderly people, and the value for things. 

I grew up amidst people who had a passion for art, science, and ancient Sanskrit literature. I was also enrolled to learn Carnatic classical music and learn slokas of Sanskrit literature. My music and Sanskrit teacher used to tell us that all the slokas, or musical compositions, have a pattern and told me to understand their pattern. Little did I know that I will end up studying pattern recognition, which is the backbone for automation and artificial intelligence 

I did my schooling in Carmel high school and loved everything about my school life. I used to participate in extracurricular activities like singing, mono-acting and inter-school competitions.  

My favorite subjects were Biology and physics. I wanted to be a scientist at that time and wanted to innovate things. I always used to think out of the box to solve a problem.  

What did you do for graduation/post graduation?

During the 12th standard, we were aware of professions – Engineering, medical, Law, and Science. While I could not get a free medical seat, I opted for engineering via the common entrance test (CET), now called (KET). I got an engineering seat in computer science at Global Academy of Technology, Bengaluru. With five to six subjects and two labs per semester, and exams every few months, the first three years of engineering flew in no time. In early 2000, few companies were coming for campus placement, my best friend N Raghavendra gave constant encouragement and helped me to prepare for interviews and I got an offer from EDS-HP as test engineering. 

After working for 1.3 years in EDS-HP, I was not satisfied with the work. I realized that an engineering degree is insufficient. I then started preparing for GATE to pursue my masters. 

I continued with my post-graduation in computer science from Dr Ambedkar Institute of Technology under Visvesvaraya technological university. I did my internship at the Indian Institute of Science (IISC) Bangalore. 

What made you choose such an offbeat, unconventional and rare career?

During my master’s, Image processing and machine learning was a hot research topic. My thesis was on flood assessment using satellite images. In 2009, there were floods in Kurnool and Mantralaya (AP) regions. My internship guide, Dr. S N Omkar, at Indian Institute of Science (IISC) asked me to develop machine learning algorithms which can automatically assess the flood region using multitemporal satellite images. I feel this was the most important turning point of my life in becoming a researcher. Since I had no knowledge in image processing and machine learning, It was a struggle. I needed to work hard, needed to read a lot of research papers, and had to learn applied mathematics to develop machine learning algorithms and even had to visit ISRO to learn about satellite images. That one year of struggle gave me a lot of knowledge in the new domain, which is now booming. We were about to submit our thesis. At that time, one of my best friends was struggling to get proper output for his thesis work on Kannada language optical character recognition. I took it as a challenge to help him as I had some confidence from the knowledge I had gained from my thesis work. We were able to solve his one-year-long project in 7 days. Solving my friend’s issue gave me the confidence that god’s grace, hard-work, determination, and dedication can lead to success. My thesis was awarded the best research paper at top international conferences. 

Tell us about your career path

My post-graduation thesis work got me into the career of machine learning and artificial intelligence. In 2010, machine learning was in the research area at the university level. Top companies like Google, Amazon, and Qualcomm started to translate machine learning research into products. I got an offer from a startup called Telibrahma Convergent Communication Private Limited after my post-graduation to develop real-time solutions using niche technologies of image processing and machine learning. 

Telibrahma was an ad delivering startup that used AI to show ads that featured on smartphones using Bluetooth and Wifi technology. 

The startup is the best place for innovation. Due to my appetite to innovate new things, i did not feel the work pressure. 

Once, we needed to show a proof of concept (POC) for a client (Ponds). To develop the POC, I worked day and night for 14 days, leading to project procurement to create a real-time application called “Age Miracle.”

“Age Miracle”, in real-time, detects the human face and automatically detects blemishes and wrinkles using image processing concepts, predicts the age of the face, and shows a clear face like in the fair and lovely advertisement. This application was deployed in kiosks in malls across India and the world, which instantly improved the sales of Ponds Age Miracle cream. 

Telibrahma had an image-based ad delivery app called “Interact.” With the increase of data day by day, the machine learning algorithm’s retraining took 24 hours. I developed a new pipeline that reduced the data dimension, which reduced the retraining time from 24 hours to 5 minutes. All these fancy things of translating academic research into real-world application happened due to constant encouragement from B S Shreyas, who led the innovation team.

My 2.5 years at telibrahma molded me to take up more responsibilities and explore more. In 2012, I moved to Robert Bosch, where I worked on vision-based algorithm development for an advanced driver assistance system (ADAS). When I joined Bosch Bangalore, I was new to the automotive domain. There was no team nor were their people working on vision-based algorithm development. My colleague S Srivathan Iyengar and I worked as an extended bench for the german team. Initially, the work was not innovative. It was like bug fixing of existing code. With our hard work and ability to learn, we gained the german team’s trust which resulted in them giving more responsible work to the Bangalore team. Within 2 years, we built a group of 15 engineers.

We got a few component responsibilities to handle, where I learned to apply image processing and machine learning algorithms in production. We even traveled to Germany to learn more about the new technology of Advanced Driver Assistance systems. The developed algorithm has been deployed on Volkswagen Passat and Land Rover cars, and assists drivers in avoiding accidents. I love Bosch as a company. It takes care of its employees and innovates things using a strong process to achieve great engineering standards. 

My 4 years in Bosch flew like anything. In September 2016, i got an opportunity to work on developing autonomous vehicles from scratch (pilot project) from the global technology office, Cognizant India. By this time, machine learning was giving way to deep- learning, with an improvement in GPU hardware and deep learning libraries like TensorFlow. The research community started to publish improved and accurate results using deep learning architecture. Again I needed to learn deep learning architecture and new coding libraries to replace the machine learning pipeline to a deep learning-based pipeline.

I single-handledly developed a driver drowsiness detection system and showcased it to senior management, which gave me the “High Impact Individual” award. This award is given to 10 people out of 220,000 employees of cognizant around the world.

I even developed a new deep learning pipeline for the traffic light and road sign recognition. I also worked on developing a driver assistance system for Indian traffic scenarios so this system could be deployed on an existing vehicle to reduce accidents and improve road safety. 

We could drive our test car without a driver on a test track using artificial intelligence. All the innovation work has been filed for patents, and few have been published in top conferences. 

During this time, I started to share knowledge with research scholars and students to help them pursue their project work in this field. By doing so,  I started to apply machine learning to a diverse fields of science.  

My childhood friend, Dr. K.L Gururajprasad, a dentist by profession, was pursuing his Ph.D. in “Smart Teaching in Orthodontics” using augmented reality. His research work needed machine learning. I started to collaborate with him to solve his research problem on orthodontics. It was a win-win situation for both of us. By helping students, I gained a lot of knowledge, made my fundamentals strong, and published many research papers.

Getting an offer abroad from India is very difficult in the field of machine learning, as companies will first look for local talent or best among the world. All the hard work gave me the opportunity to work in one of the top research centers of the world called A*STAR Singapore (Agency for science, technology, and research) 

How did you get your first break?

Due to my post-graduation thesis work, I got an offer from a startup where I started working on image processing and machine learning.

What were the challenges you faced in your career? How did you address them?

Challenge 1: When I was studying 10th standard, my father’s company closed down suddenly which limited our financial backup. I needed to pursue my higher studies. By putting restrictions on spending, I was able to mitigate the financial crisis.

Challenge 2: After my engineering, I was working in EDS-HP. I was not satisfied with the work and realized that I had to pursue post-graduation. My mother was hesitant that I was leaving my job for post-graduation as I was a sole earning person in the family and my parents had some minor health issues. My father’s bold decision and my determination to pursue post-graduation have given a significant dividend in my life. 

Where do you work now? Tell us about your work

Presently, I am working as a Senior Research Officer at Singapore Bioimaging Consortium (SBIC) A*STAR, Singapore. I am working in the Signal and Image Processing Lab, headed by Dr. K N Bhanu Prakash. I am involved in developing new deep learning techniques for Biomedical Imaging. 

The skills needed for this job are research mindset with in-depth knowledge in machine learning, deep learning, applied mathematics, and statistics. I acquired all the above expertise right from my post-graduation by solving real-world problems and constantly upgrading my skill sets. 

What is a typical day like?

I have the habit of getting up early, by 5.30 am. Practice yoga for one hour and prepare food and reach the office by 9 am. I generally plan the day’s work or have a working road map with my reporting officer for one week.  I work till 6.30 pm and hit the gym for 1 hour and attend few workout sessions like hatha yoga and LesMills and burn. I read research manuscripts or try to listen to TED talks while traveling in the metro. I try to follow the early raise and early bed principle. But if there is any deadline or urgent work, I do night outs. 

What is it you love about this job? 

Presently, I am working on developing deep learning algorithms for pre-clinical and clinical data. I need to find novel solutions to get accurate results. There is scope for learning, innovation, and spinning of a new startup based on our research achievements. 

How does your work benefit society? 

Medical diagnosis requires precise, accurate detection, prediction, and classification of medical issues using biomedical images.  There is a need for artificial intelligence (AI) to assist medical personnel in cases where there is a lack of a radiologist or lab technician, especially in emergency scenarios where early diagnosis, prognosis, and early medication can cure or save a patient’s life. 

Deep learning architectures have proven to provide more accurate results than humans in detecting cancer regions, segmentation of fat deposits, and understanding heart scars. 

One such research that will help the present generation is obesity analysis. Children and mid-age adults are getting obese due to lack of exercise and a sedentary lifestyle. Obesity is called the mother of all diseases. Obesity leads to type 2 diabetes, heart-related issues, and even leads to cancer. Covid fatality is due to the high lipid content in the human body. 

Recent research has proven that BMI (Body Mass Index) is not the correct way of understanding human fat. The best way to understand the human fat composition is via Computerized Tomography (CT) and Magnetic Resonance Imagining (MRI). Using deep learning segmentation algorithms, we can automatically identify the different fat percentages in each region and predict prognosis. This will aid the doctors in better medication to save/cure/ help the patients.  

We have developed deep learning/ artificial intelligence tools /pipelines which will help healthcare workers understand the medical condition faster during an emergency, thus saving a human’s life. It also reduces human error and aids in accurate diagnosis of the disease 

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

Few notable works are close to me. 

  1. I worked in automatic self-glare detection in a vision-based driver assistance system. The algorithm developed by reading the research paper was flashed into car ECU (electronic component unit). The self-glare functionality worked perfectly when tested on-road in real time on autobahn of Germany. 
  1. The deep learning algorithm pipeline was developed for a real-time drone-based automatic spray of pesticide which monitored the greenhouse precisely. This work got the outstanding paper award at a top international conference.  

Your advice to students based on your experience?

Keep on learning, choose a field which you have passion for. Always keep a positive attitude—the secret of success is, good attitude,  hard work, consistency, and learning from failures. 

Future Plans?

I want to continue to work in applied research and give my best to solve real-world problems.