Revolutionising industries with AI: Sumedh Datar's journey in deep learning and computer vision

Beyond his technical accomplishments, Datar is dedicated to educating others in his field

By Ammar Tarique

  • Follow us on
  • google-news
  • whatsapp
  • telegram

Top Stories

Published: Tue 5 Dec 2023, 5:25 PM

Artificial Intelligence techniques, such as deep learning, have garnered significant attention in 2023, especially with the emergence of platforms like ChatGPT making headlines. Recently, the US, under the Biden administration, has supported AI initiatives by passing an executive order. Companies, including those behind Falcon models from the Gulf, are not only investing heavily in these technologies but are also making substantial contributions to the global landscape. They have recognised the crucial role and transformative impact of this technology, advancing with the latest open-source models that benefit the entire world. This technology extends far beyond simply answering questions and creating written content.

Deep learning has revolutionised many fields, from improving medical diagnosis times to increasing efficiency in retail — and one man who has been at the heart of these innovations is Sumedh Datar.

A quick background

Datar is a senior machine learning engineer who specialises in deep learning solutions and computer vision. The former uses algorithms to notice patterns, and make predictions with data; meanwhile, computer vision can interpret images and videos.

Combining the two has the potential for powerful applications that go beyond text-based tools like ChatGPT, as Datar’s career achievements prove.

Changing the medical field for the better

Datar began his journey with a bachelor's in medical electronics, followed by a master's in computer science. The combination of these fields was key to his success.

In his initial role as a computer vision engineer, Datar and his team identified issues with existing cervical cancer detection methods. Traditionally, experts manually examined slide images under a microscope, a process taking weeks or months and prone to errors. To address this, Datar employed deep learning techniques for more efficient and accurate cervical cancer identification.

He utilised a small, labeled dataset from manual methods and applied transfer learning, a deep learning variant, to this data, testing it on numerous new images. This innovative approach significantly improved accuracy, achieving a 20 to 25 per cent increase over the baseline metric. The application he developed, now used by various healthcare providers and tested on over 800 individuals, has enhanced treatment quality. Datar's crucial contribution notably raised his company's profile and earned widespread recognition in the healthcare industry.

"I was surprised to see the deep learning algorithm's results, where the model identified patterns with near-human accuracy," Datar remarked.

These advancements not only enhance diagnostic accuracy but also expedite diagnosis times, which can be life-saving. By reducing healthcare costs, deep learning algorithms also improve access to care, especially beneficial for people in underserved regions.

From healthcare to retail

Datar's journey didn't end with healthcare.

In his subsequent role, he transitioned to the retail industry, continuing his focus on deep learning. Working for a major retail convenience brand, he observed that resource limitations were hindering customer service and store efficiency. The challenge was enabling self-checkout in a setting where many products lacked identity, making traditional methods like bar codes impractical. Datar saw computer vision as the solution which uses visual features like colour, packaging etc.

Addressing this issue was initially challenging due to data scarcity, but Datar's innovative approach led to the development of a product recognition model. This solution allowed stores to operate efficiently, even with a limited number of staff during significant frontline worker shortages. It enhanced customer experience, reduced operational costs, and prioritised safety by encouraging digital payments. His algorithm for identifying unique products achieved an accuracy rate in the high 90s. Datar, holding over 10 patents in the convenience retail sector and an international patent in identity management, has made significant contributions to the field.

"Computer vision algorithms, once mastered, are incredibly user-friendly and have a profound impact, changing lives worldwide," says Datar.

What does the future hold?

Despite his significant contributions to the medical and retail sectors, Datar aspires to make an even broader impact. He plans to leverage his skills in engineering, research, and product development to create applications and algorithms that not only revolutionise industries but also enhance people's lives.

Beyond his technical accomplishments, Datar is dedicated to educating others in his field. He has assisted numerous individuals in securing data science positions. Additionally, he has shared his insights through expert talks, including engagements with the Government of India, and is recognised as an accomplished speaker.

Datar is a published author in outlets like the Deccan Herald and Hackernoon and holds a fellowship with the prestigious British Computing Society.

— Ammar Tarique is a business journalist.

More news from KT Network