Kaushal Chapaneri
Machine Learning Engineer
Python, Remote Sensing, Image Processing
Kaushal Chapaneri is a passionate developer with extensive experience in building Machine Learning base solutions. He is a strong believer of continuous learning and development. He had done his Bachelor’s Degree in Computer Science in 2016. He had done his Master of Technology in Computer Science from Dharmsinh Desai University. He had done his research work on AI for Climate change at ISRO. He has a deep knowledge of Python, Machine Learning, Deep Learning, Remote Sensing and Image Processing. He says, Remote Sensing + Image Processing + Machine learning together can solve many real world problem of the world. He has done multiple online courses on Machine Learning, Deep Learning from various online platforms like DataCamp, Udemy and NVIDIA. He reads articles and blogs regularly to update himself. He works on Kaggle to understand many new codes. He has done many online courses to get new technology. Here he shares his interest, experience and vision with us.
Q1. Please tell us about your journey up to become machine learning engineer?
My first encounter with Machine learning was during my bachelor’s degree, and I was fascinated by how a machine can learn by itself and their ability improves over time. Later I joined the master’s program in computer science focused on machine learning. And it leads me to join ISRO as a researcher for my thesis work. At present, I’m working as a Machine learning engineer at a product based startup called Intellica.ai
Q2. What was your focused area during research work at ISRO?
My research topic was AI for Climate change, focused on Oceanography. To be specific, how we can tackle Oil-spill pollution problem in oceans using machine learning. Every year tons of gallons of oil get released in ocean, which affects marine environment as well as mankind. Further it triggers fire hazard which leads to air pollution. To detect oil spills from satellite imagery we developed semi-supervised technique which can detect and classify oil spills from images effectively and gives size of affected area.
Q3. How Remote Sensing and Image Processing help to solve the problem?
Remote Sensing gives ability monitor an object without making direct contact with it, and it provides valuable data over vast area in short time. After capturing an image, it requires certain processing steps in order to use it further and derive insights from it. So image processing is must for remote sensing. I would say Remote Sensing + Image Processing + Machine learning helps to solve the real world problem. Together they provide solution in field of Agriculture, Geology, Forestry, Ocean & Costal monitoring etc.
Q4. Is really Machine Learning changing the world?
Definitely Machine learning is changing the world. It is bringing advancement in each and every sector. Nowadays all business is moving towards providing AI enabled solution. Cutting edge research work in Medical, Space exploration, Automobiles etc is possible due to Machine learning.
Q5. Which type of machine learning you have used maximum in your professional work?
It totally depends on problem we are solving, what type of data we have and what is the expected outcome. I have used supervised approach majority of the time but semi supervised and unsupervised is also used in some projects.
Q6. . How participating in ML hackathons is different than developing a ML based product?
ML hackathons are great for gaining practical knowledge and learn new things from peers at the same time but its scope is limited up to “submission.csv” , wherein product development process one has to keep many performance factors in mind like size of model, inference time of model, feasibility of model integration etc. ML engineer has to develop APIs that interacts with model and generates predictions from it, later they need to pass on those model results to front-end development team in order to display on client side.
Hackathons has comfort zone, in which everything happens in prescribed manner, like what type of input will be feeded to model, what will be size of it etc. In case of product scenario is totally opposite, user can input anything irrespective of instruction given, so in such cases model will not be able to generate prediction.
Q7. You have done multiple online certificate courses. Do these courses really help and you recommend to beginners to first go for those course?
I would say online courses gets you started and puts on path from where you can develop something useful based on learning from those courses. Nowadays video lectures of IITs, MIT and Stanford’s available on internet, one can definitely go for it and also do the online certification course.
Q7. How you learn new things and keep yourself updated?
I read articles and blogs on daily basis, sometimes attend webinars or listen to AI/DS podcasts. I read kernels on kaggle and try to understand code. I also go for online courses to learn new things.
