Hello Somdip, Can you please introduce yourself?
Hello and Namaste! I am Somdip Dey, an Embedded AI Scientist at the University of Essex, UK, affiliated with the Embedded and Intelligent Systems Laboratory, and a researcher at the Samsung R&D Institute, UK. My work involves developing Artificial Intelligence (AI) based solutions on embedded devices such as mobile phones, which are cost-effective and affordable.
What are your focus areas and why?
I was born in Kolkata, India and have spent most of my childhood and adolescence in India. Growing up in India we faced so many issues such as food waste, malnutrition, pollution, crazy traffic, which could have been better improved using technology. After moving to the UK and with access to more resources in terms of technology I realized that there is so much I could do to improve lives with the use of such technology. Henceforth, I started utilizing my knowledge to develop affordable traffic categorization solutions and applications to reduce food waste, all powered by AI in the backend.
How is your work different from the trend?
At the moment the majority of AI-based solutions available are processed off the device and, on the cloud, and/or on the servers or mixture of both. AI solutions are very demanding in terms of computing resources and embedded devices are not usually used for data processing, rather used for data collection. Although partially the data could be processed on the embedded device, most of the data is sent to the cloud or to the servers via internet for further processing and computation to produce effective results. The issue with these types of approaches is that for developing countries where accessing the internet in many places is still a big challenge and the associated astronomical cost of hosting servers could discourage such developing countries to adopt AI-based solutions. Moreover, many available popular AI-based solutions heavily depend on the data for training and hence, depending on the data the solution could change. Hence, having a flexible solution, which is capable of adapting to the data, is also necessary.
My work involves developing AI-based solutions on the devices so that processing and computation are all done on the device without the need to send the data to the cloud and/or the servers. However, such an approach has its own limitations. My research is focused on making AI solutions lightweight so that they could be computed on the device without increasing the power consumption and overheating the device itself. In 2018, I developed a traffic categorization solution, which is capable of executing completely on the device and adapts to its surrounding like a human being to learn from it. This work has been presented at the 2018 NASA/ESA Conference on Adaptive Hardware Systems (NASA/ESA AHS) and was a great opportunity to be able to share my work with the research community.
How do you see the future of “Artificial Intelligence” in the next 5 years?
We can already notice the use of AI in every industry in some form or other. Although at the moment most of the solutions available are in the development phase, we could certainly see the emergence of more mature applications of AI in the coming 5 years. I am also hopeful that more and more researchers would be developing AI applications, which are affordable not just by developed countries but by developing countries as well.
Living in London, I am really grateful to meet so many like-minded people because of the multi-cultural environment London has to offer. So, I hope London would play an important role in innovations in the AI field in the coming few years and improving lives in general.
The last work or final thoughts for our readers?
I am aware that many people are embracing AI’s application as a good change, whereas, many are also skeptical to adopt such a change. I could only say that AI is here to help make our lives better and not worse. So, having an open mind to adopt such change is necessary for AI’s further development. Additionally, if you are a researcher in the field and want to join the movement of making AI affordable so that it is accessible by anyone, please do drop me a message on somdipdey.co.uk and/or tweet me @somdipdey.